View in EDS HTML Full Text PDF Full Text

The Rise in Occupational Coding Mismatches and Occupational Mobility, 1991-2020

Saved in:
Bibliographic Details
Title: The Rise in Occupational Coding Mismatches and Occupational Mobility, 1991-2020
Language: English
Authors: Andrew Taeho Kim (ORCID 0000-0002-1402-2927), ChangHwan Kim (ORCID 0000-0001-7149-1386)
Source: Sociological Methods & Research. 2026 55(2):659-699.
Availability: SAGE Publications. 2455 Teller Road, Thousand Oaks, CA 91320. Tel: 800-818-7243; Tel: 805-499-9774; Fax: 800-583-2665; e-mail: journals@sagepub.com; Web site: https://sagepub.com
Peer Reviewed: Y
Page Count: 41
Publication Date: 2026
Document Type: Journal Articles
Reports - Research
Descriptors: Occupational Mobility, Coding, Occupations, National Surveys, Reliability
Assessment and Survey Identifiers: Current Population Survey
DOI: 10.1177/00491241241303517
ISSN: 0049-1241
1552-8294
Abstract: Occupation is a construct prone to classification mismatches by coders and description inconsistency by respondents. We explore whether mismatches in occupational coding have recently increased, what factors are associated with the rise in mismatches, and how the rise affects estimates of intragenerational occupational mobility. Utilizing the 1991-2020 Annual Social and Economic Supplement of the Current Population Survey, which collects information on respondents' current occupation and the previous year's main occupation, we identify coding mismatches and compare the probabilities of occupational mobility based on four combinations of two variables. Our results show that not only do the estimates of occupational mobility between two adjacent years vary substantially across measures, but also that the magnitudes of intragenerational occupational mobility across measures become increasingly decoupled over time. We demonstrate that the likely cause of this divergence is the rise in coding mismatches between coders. We discuss the implications of our findings.
Abstractor: As Provided
Entry Date: 2026
Accession Number: EJ1502110
Database: ERIC
Full text is not displayed to guests.
FullText Links:
  – Type: pdflink
    Url: https://content.ebscohost.com/cds/retrieve?content=AQICAHj0k_4E0hTGH8RJwT4gCJyBsGNe_WN95AvKlDbXJGqwxwGx7WBMscCVr8r6e9kpDDzsAAAA4zCB4AYJKoZIhvcNAQcGoIHSMIHPAgEAMIHJBgkqhkiG9w0BBwEwHgYJYIZIAWUDBAEuMBEEDCg90XrizZCVlgiMNgIBEICBmwgga6ZSyVAMQPOWO1nXuLp-PeClPe_gZNi8RSPJ6O57-pf2oOYVYWBDN8K_vGcdWAi5hVsW-I_N6dYmYrAk_OhFprl27P-OWAI1x95Zr2U5Dy_tKguaN6qRae0duBFQjB4g_JmRJvosiUsGp07Bejl-QvVErJ5CWU9rO9seFd5CFTwb665IZdzwg5OthkGpEf02iTHak0s6uQ7A
Text:
  Availability: 1
  Value: <anid>AN0192656321;som01may.26;2026Apr02.02:51;v2.2.500</anid> <title id="AN0192656321-1">The Rise in Occupational Coding Mismatches and Occupational Mobility, 1991–2020 </title> <p>Occupation is a construct prone to classification mismatches by coders and description inconsistency by respondents. We explore whether mismatches in occupational coding have recently increased, what factors are associated with the rise in mismatches, and how the rise affects estimates of intragenerational occupational mobility. Utilizing the 1991–2020 Annual Social and Economic Supplement of the Current Population Survey, which collects information on respondents' current occupation and the previous year's main occupation, we identify coding mismatches and compare the probabilities of occupational mobility based on four combinations of two variables. Our results show that not only do the estimates of occupational mobility between two adjacent years vary substantially across measures, but also that the magnitudes of intragenerational occupational mobility across measures become increasingly decoupled over time. We demonstrate that the likely cause of this divergence is the rise in coding mismatches between coders. We discuss the implications of our findings.</p> <p>Keywords: occupation; coding mismatch; occupational mobility; current population survey</p> <hd id="AN0192656321-2">Introduction</hd> <p>As stressed by [<reflink idref="bib62" id="ref1">62</reflink>], the categorical nature of inequality is a well-documented idea in research on labor market inequality. One variable that represents that categorical nature is occupation. Occupation has been the main sociological variable in studying socioeconomic attainments as well as inter- and intragenerational mobility ([<reflink idref="bib8" id="ref2">8</reflink>]; [<reflink idref="bib64" id="ref3">64</reflink>]). Most class theories use occupation as a way to group people into different class categories, at least in combination with other variables ([<reflink idref="bib16" id="ref4">16</reflink>]; [<reflink idref="bib68" id="ref5">68</reflink>]; [<reflink idref="bib67" id="ref6">67</reflink>]). Recently, changes in occupation within one's lifetime have gained much sociological attention ([<reflink idref="bib11" id="ref7">11</reflink>]; [<reflink idref="bib30" id="ref8">30</reflink>]; [<reflink idref="bib32" id="ref9">32</reflink>]). These studies use occupation at different points in time to measure within-individual changes in their positions in social strata. One of the key assumptions behind the extensive usage of occupation in sociology over other socioeconomic measures such as income is its validity and reliability in measurement ([<reflink idref="bib21" id="ref10">21</reflink>]; [<reflink idref="bib64" id="ref11">64</reflink>]). However, there is a healthy debate about how accurate and reliable occupation is as a variable for measuring status or for use in studies of inequality and mobility (e.g., [<reflink idref="bib5" id="ref12">5</reflink>]; [<reflink idref="bib12" id="ref13">12</reflink>]; [<reflink idref="bib26" id="ref14">26</reflink>]; [<reflink idref="bib34" id="ref15">34</reflink>]; [<reflink idref="bib38" id="ref16">38</reflink>]; [<reflink idref="bib42" id="ref17">42</reflink>]; [<reflink idref="bib46" id="ref18">46</reflink>]; [<reflink idref="bib54" id="ref19">54</reflink>]).</p> <p>Contrary to widely accepted assumptions in sociology, multiple studies have demonstrated that occupation is a construct prone to classification mismatches and low measurement validity (e.g., [<reflink idref="bib5" id="ref20">5</reflink>]; [<reflink idref="bib34" id="ref21">34</reflink>]; [<reflink idref="bib38" id="ref22">38</reflink>]; [<reflink idref="bib48" id="ref23">48</reflink>]; [<reflink idref="bib49" id="ref24">49</reflink>]). For example, studies using the US datasets have reported that when two professional coders evaluate the same descriptions of jobs and occupational activities, they tend to agree on which occupational code to assign only about half the time (e.g., [<reflink idref="bib34" id="ref25">34</reflink>]; [<reflink idref="bib48" id="ref26">48</reflink>]; [<reflink idref="bib49" id="ref27">49</reflink>]). The lack of reliability in occupational coding is not independent of the reliability problem in occupational mobility. [<reflink idref="bib26" id="ref28">26</reflink>] showed that the use of occupation in intergenerational mobility tables is prone to classification mismatches and measurement errors, the latter stemming from proxy reporting where a respondent answers survey questions on the occupation of other family members. As for intragenerational mobility, in a study using the 1976–2004 Current Population Survey (CPS), [<reflink idref="bib34" id="ref29">34</reflink>] presented that annual occupational mobility had increased when two independently coded occupations were used while month-to-month mobility, which relies on the dependent coding of occupation, did not increase. This divergence between the two mobility measures implies that the reliability of the occupational coding may have declined further over time.</p> <p>To our knowledge, no previous study has explored whether the likelihood of coding mismatches has changed over time. The previous study of occupational mismatches in the CPS covered the period up to the early 2000s ([<reflink idref="bib34" id="ref30">34</reflink>]). Considering that non-response rates to earnings in the CPS have increased noticeably over the last several decades and that one of the key variables in imputing earnings is occupation ([<reflink idref="bib25" id="ref31">25</reflink>]), the reliability of occupational coding has become more important than ever.</p> <p>Using the Annual Social and Economic Supplement of the Current Population Survey (CPS-ASEC), this study explores the extent to which coding mismatch between coders can bias our understanding of occupational mobility and tracks changes in coding mismatches and occupational mobility over time. There are two ways to gather an individual's occupation from two adjacent years in the CPS-ASEC. The first is simply to use and compare two occupation variables in the same March survey: the main occupation in the previous year and the current occupation. These two occupations are coded in reference to each other, thus dependently coded. The second is to take advantage of the rotating panel nature of the CPS by linking the same individual across two years so that we can compare individuals' current occupations in two adjacent years. Unlike the first method, two current occupations in adjacent years are likely to be coded by two independent coders without referring to the previous occupation, thus they are independently coded. Exploring the differences between the two methods is the core of this project. In principle, two methods should yield similar, if not identical, results as long as occupations are measured reliably. However, as we will demonstrate later, the two results have substantial differences. In addition, we examine whether the rise in intragenerational occupational mobility reported in previous studies ([<reflink idref="bib30" id="ref32">30</reflink>]; [<reflink idref="bib33" id="ref33">33</reflink>]; [<reflink idref="bib37" id="ref34">37</reflink>]) is consistently observed between the two methods or depends on the way the occupation question is coded. This study sheds new light on the reliability of occupation variables and the implications of coding mismatches for the study of mobility.</p> <hd id="AN0192656321-3">Literature Review</hd> <p></p> <hd id="AN0192656321-4">Coding Occupations</hd> <p>In surveys such as the Decennial Censuses, the CPS, and the General Social Survey (GSS), respondents do not answer directly with the title of their occupation. Rather, they describe their workplace activities and professional coders assign an appropriate code. The CPS asks two questions: "What kind of work do you do, that is, what is your occupation" and "What are your usual activities or duties at this job." This approach is preferred to asking respondents to choose their own occupational category because classification systems such as the Standard Occupational Classification of the US Bureau of Labor Statistics are often too complex for non-experts ([<reflink idref="bib29" id="ref35">29</reflink>]). The ease of answering questions about occupation lies not in knowing the title of one's occupation, but in being able to describe what one does.</p> <p>There are two potential problems with this practice. First, there is no guarantee that respondents' descriptions of duties and activities are accurate and sufficient. When performing multiple tasks, people tend to emphasize the more prestigious part of their tasks, thus resulting, for example, in a higher proportion of managers ([<reflink idref="bib24" id="ref36">24</reflink>]; [<reflink idref="bib38" id="ref37">38</reflink>]). This is related to the validity of occupation. The assumption that occupation questions are easy to answer implies that respondents provide consistent descriptions of their occupation and job duties, and thus, the reliability of respondents' descriptions (let's call it <emph>response reliability</emph>) is high. The validity of occupation assumes high response reliability. Second, coders may have different opinions on which occupational codes should be assigned, which challenges <emph>intercoder reliability</emph>. In essence, measurement problems with occupation can occur at two stages: first, at the recall stage by respondents, which raises a question of validity and response reliability, and second, at the in-office processing stage by the coders, which raises a question of intercoder reliability. All survey data, even variables that do not require the assignment of code in the office, face issues at both stages ([<reflink idref="bib4" id="ref38">4</reflink>]). However, it can be reasonably assumed that the error at the second stage is relatively small for most variables ([<reflink idref="bib7" id="ref39">7</reflink>]). In sociology, occupation is often preferred to income as a measure of socioeconomic status because the likelihood of error at the first stage is presumed to be small (= assumption of high response reliability and high validity) and the problem at the second stage is not considered a concern (= assumption of high intercoder reliability) ([<reflink idref="bib21" id="ref40">21</reflink>]; [<reflink idref="bib64" id="ref41">64</reflink>]).</p> <p>Unlike other variables, however, the measurement problems in the second stage are not negligible for occupation. Occupational classification is socially constructed, and there is no classification that is free from subjectivity ([<reflink idref="bib38" id="ref42">38</reflink>]). Most job duties and activities involve a mix of various activities, making it necessary to group heterogeneous jobs into the same occupational categories ([<reflink idref="bib63" id="ref43">63</reflink>]). Indeed, by analyzing the verbatim text of the GSS, [<reflink idref="bib46" id="ref44">46</reflink>] showed substantial variations in descriptions within the same occupation. Occupational classification schemes assume that occupations are mutually exclusive, but many jobs involve tasks spanning multiple occupations ([<reflink idref="bib24" id="ref45">24</reflink>]). Thus, disagreement between coders per se is not an error but a reflection of the socially constructed nature of the occupation variable. At the same time, agreement between coders does not guarantee that the occupational coding is valid and always reliable.</p> <p>Previous studies have demonstrated that mismatches of occupation codes between coders are quite common ([<reflink idref="bib9" id="ref46">9</reflink>]; [<reflink idref="bib12" id="ref47">12</reflink>]; [<reflink idref="bib14" id="ref48">14</reflink>]; [<reflink idref="bib38" id="ref49">38</reflink>]; [<reflink idref="bib44" id="ref50">44</reflink>]). According to the review by [<reflink idref="bib44" id="ref51">44</reflink>], the mismatch rates for detailed occupational categories range from 11 percent to 56 percent. Using the 1997–1998 CPS occupation descriptions, [<reflink idref="bib12" id="ref52">12</reflink>] examined the match rate of two coders at the three-digit level and found that the mismatch rate was 29 percent. The usage of broad occupational categories does not eliminate the mismatches. When coders disagreed, around half of the mismatches occurred at the one-digit level ([<reflink idref="bib12" id="ref53">12</reflink>]). The low intercoder reliability has also been commonly reported in surveys in other countries. For example, [<reflink idref="bib5" id="ref54">5</reflink>] discovered that the disagreement rate between two coders for one-digit occupational codes was about 28 percent–30 percent in a Dutch sample. In a recent study using the Korean General Social Survey, [<reflink idref="bib38" id="ref55">38</reflink>] reported a 31 percent mismatch rate for one-digit occupations and 51 percent for three-digit occupations. Interestingly, the lengthy description of duties and activities leads to a higher rate of mismatches between coders rather than reducing them ([<reflink idref="bib12" id="ref56">12</reflink>]; [<reflink idref="bib38" id="ref57">38</reflink>]). Assuming that the lengthy description increases the validity of the occupation, the validity of the occupation measurement is negatively associated with the intercoder reliability of occupational coding. This may be because the lengthy description lowers response reliability. As respondents provide more detailed information on their occupations and job duties, the likelihood of discrepancy between the two descriptions may increase. The mismatch of the occupation codes between the two coders is also related to respondents' characteristics. Mismatch rates tend to be higher among the less-educated, older respondents, and lower-income earners than their counterparts ([<reflink idref="bib38" id="ref58">38</reflink>]).</p> <p>Regarding changes in occupational validity and intercoder reliability over time, there are reasons to believe that they have recently declined in the CPS. Since 2010, the non-response rate of the CPS has more than doubled from 6 percent to 15 percent ([<reflink idref="bib6" id="ref59">6</reflink>]). Even among those who participate in the CPS, the item non-response rates such as earnings have grown significantly ([<reflink idref="bib39" id="ref60">39</reflink>]). These trends indicate that respondents are less likely to provide information about their labor market activities, although the non-response rate for occupation has not increased.[<reflink idref="bib7" id="ref61">7</reflink>] The rapid transitions in occupational structures and job activities could result in a rise in coding mismatches as well. On the other hand, we cannot rule out the possibility that selective item response rates and changes in the composition of the respondent population, such as an increase in the highly educated, could lead to a higher match rate.</p> <hd id="AN0192656321-5">Occupational Coding Mismatches and Occupational Mobility</hd> <p>The high mismatches between coders can bias the outcomes even when the discrepancies between coders are completely random. The mismeasured dependent variable tends to affect only statistical precision, while measurement errors in independent variables cause biases in estimation ([<reflink idref="bib22" id="ref62">22</reflink>]; [<reflink idref="bib52" id="ref63">52</reflink>]). However, in the case of dichotomous dependent variables such as job changes, mismeasurement can lead to biased and inconsistent estimates ([<reflink idref="bib31" id="ref64">31</reflink>]). One such case is occupational mobility. When occupations in two different years are coded by independent coders, the estimated occupational mobility can reflect actual changes, inconsistent descriptions of occupations by respondents, and disagreement between coders. The last two sources are likely to lead to an upward bias in the estimate of occupational mobility.</p> <p>Disagreement between coders stems from two sources. The first is low intercoder reliability, where coders assign different codes to the same description. The second arises from alterations in respondents' descriptions. Even with high intercoder reliability, subtle variations in descriptions of the same occupation can lead to false mobility when coding is performed independently. Suppose that respondents who remain in the same occupation provide slightly different descriptions of their occupation and job duties in multiple surveys. Professional coders who do not know that the same respondents provided these descriptions may agree to assign them to different occupational codes. For example, according to the Census Bureau's Standard Occupational Classification (SOC),[<reflink idref="bib8" id="ref65">8</reflink>] painters are coded 6410, which is part of the construction occupation, and painting workers are coded 8810, which is a production occupation. <emph>The Alphabetic Indexes of Industry and Occupation</emph> guide that if the painter mentions "house," it is 6410, otherwise, 8810. If the respondent mentions "landscape," then it should be coded as 2610, an artist ([<reflink idref="bib10" id="ref66">10</reflink>]).[<reflink idref="bib9" id="ref67">9</reflink>] For another example, a Certified Public Accountant working as a financial analyst in a university center for research can be coded either "0800 Accountants and auditors" or "0845 Financial and investment analysts" depending on how she describes her job duties. Professional coders are instructed to assign different codes to these slightly different descriptions, and they may agree with each other. In these cases, respondents who do not change occupations are falsely marked as occupational movers. This is a unique characteristic of the socially constructed nature of occupational coding. The validity of occupational coding is assumed when multiple professional coders agree on the coding (= intercoder reliability), but intercoder reliability does not guarantee that the measure of occupational mobility is valid. Thus, when two occupational descriptions are independently coded, coder disagreement in occupational mobility encompasses both the issues of intercoder reliability and response reliability.</p> <p>Recent studies show that occupational career mobility in the United States has increased ([<reflink idref="bib30" id="ref68">30</reflink>]; [<reflink idref="bib37" id="ref69">37</reflink>]). Although these studies rely on panel datasets, they are not completely free from the problem of coding mismatches. In analyzing the 1968–1980 Panel Study of Income Dynamics (PSID), [<reflink idref="bib34" id="ref70">34</reflink>] measured two occupational mobility rates: one based on the original occupation and the other based on the occupation reported in the Retrospective Occupation Industry Files of the PSID. They found that the occupational mobility measured using the Retrospective File (14 percent) was much lower than that based on the original occupational code of the PSID (27 percent) ([<reflink idref="bib34" id="ref71">34</reflink>]: 176–177). They attribute the difference to the fact that the coder has access to occupational descriptions from all years when coding retrospective occupations, which is dependent coding. In contrast, the original file coders only had access to the description of the respondent's current occupation, which is an independent coding. This result suggests that dependent coding can be an answer to both intercoder reliability and response reliability problems.</p> <p>In the same paper, [<reflink idref="bib34" id="ref72">34</reflink>] assessed the difference in annual occupational mobility between two ways of measuring mobility in the CPS: one based on current occupations asked of the same respondents in two adjacent years and the other based on previous occupation and current occupation asked in the same year. They reported that the former is astonishingly high at ∼40 percent, while the latter is only one-sixth of the former (p. 182).[<reflink idref="bib10" id="ref73">10</reflink>] They also presented that dropping imputed occupations after 1988 significantly affected the estimation of occupational mobility and that monthly occupational mobility in three-digit occupations of the CPS declined between 1994 and 2004.</p> <p>The reliability problem of occupation, however, is not a concern for all researchers. In defending the usage of occupation variables, [<reflink idref="bib27" id="ref74">27</reflink>] argued that the reliability of occupational coding is high. [<reflink idref="bib27" id="ref75">27</reflink>]'s estimation is based on occupational rankings rather than occupational categories. Occupational rankings measured by prestige, education, and income seem to be associated with smaller heterogeneity in job titles ([<reflink idref="bib46" id="ref76">46</reflink>]) and could be more reliable than categorical classifications ([<reflink idref="bib1" id="ref77">1</reflink>]; [<reflink idref="bib38" id="ref78">38</reflink>]). Probably aware of the reliability problem of occupational categories, some recent assessments of occupational mobility use percentile ranks rather than discrete occupation codes ([<reflink idref="bib58" id="ref79">58</reflink>]; [<reflink idref="bib59" id="ref80">59</reflink>]). However, it is not well understood how effective such an approach is in reducing the bias caused by high coding mismatches. In our empirical analyses, we test whether the gradational occupational ranking can solve the reliability problem of occupation in measuring occupational mobility.</p> <hd id="AN0192656321-6">Four Measures of Occupational Mobility of the Linked CPS-ASEC</hd> <p>In this study, we aim to estimate the potential impact of coding mismatches between coders on occupational career mobility using the CPS-ASEC. The CPS is the main data resource to study labor market activities and income inequality in the United States. For the CPS, there are basic monthly surveys and the Annual Social and Economic Supplements in March (i.e., CPS-ASEC).[<reflink idref="bib11" id="ref81">11</reflink>] In March, the same respondents answer both the basic monthly survey and the March supplement. To assess occupational mobility, we exploit the unique mini-panel design of the CPS. The sampled households of the CPS are surveyed for four consecutive months, rotate out of the sample for the next eight months, and are surveyed again for the next four months. The same respondents appear in two adjacent years in this rotation sample. Thereby, the respondents of the CPS-ASEC appear in the March surveys of two consecutive years. Using this aspect, we can build a mini-panel dataset of the CPS. For a given <emph>t</emph> year CPS-ASEC, the data can be linked forwardly (i.e., to the <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>+</mo><mn>1</mn></math> </ephtml> year) or backwardly (i.e., to the <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> year), depending on when the CPS households are sampled. For example, for about half of the 2010 CPS-ASEC respondents, 2010 was their first survey year, and they were surveyed again in 2011. Thus, their 2010 responses can be forwardly linked to their 2011 responses. For the other half of respondents, 2010 was their second year, and they were first surveyed in 2009. These 2010 respondents can be backwardly linked to their 2009 responses. Note that respondents are not surveyed three years in a row and appear only in two adjacent years.</p> <p>All CPS surveys ask about the respondent's current occupations as part of the core questions. In addition to this question, the CPS-ASEC asks about the respondents' main occupation from the previous year. We call the former occupation in the CPS-ASEC <emph>current</emph> occupation ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> ) and the latter <emph>previous</emph> occupation ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> ). <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> are not necessarily coded by the same coder but are coded dependently. We can estimate occupational mobility by comparing the previous occupation to the current occupation. The comparison between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> is not the only way to estimate occupational mobility between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and <emph>t</emph>. Once the CPS-ASEC mini-panel is constructed, the change in the two current occupations between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and <emph>t</emph> can be compared (i.e., <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> vs. <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> ). These two current occupations are answered by the same respondents in two adjacent years. Another way to estimate occupational mobility is to compare the previous main occupation in year <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> with the previous main occupation in year <emph>t</emph> (i.e., <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> vs. <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mi>t</mi></msub></math> </ephtml> ). Two previous occupations are also answered in two adjacent years. Thus, there are three ways to estimate occupational mobility between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and <emph>t</emph> using the CPS: comparing the previous occupation to the current occupation ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> vs. <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> ); comparing two current occupations ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> vs. <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> ); and comparing two previous occupations ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> vs. <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mi>t</mi></msub></math> </ephtml> ).</p> <p>In the basic monthly surveys, respondents are asked whether their employers and job activities have changed from the previous month so the month-to-month changes in occupations are conditional on such changes. When respondents report no change in job and employer from the previous month, their occupations are not changed. This is called dependent coding. Only when employers and job activities are changed between month-to-month surveys, the current occupations are coded independently. The CPS introduced dependent coding for month-to-month changes in 1994. However, in the fifth month, when respondents are reinterviewed after the eight-month break, the questions about job and employer changes are not asked, and thus <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> coded independently without referring to <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> . Like two current occupations, two previous occupations, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mi>t</mi></msub></math> </ephtml> , are independently coded. Contrary to this, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> , which are asked in the same survey, are coded dependently. Thus, among the three measures of occupational mobility, the comparison between the previous occupation and the current occupation asked in the same month in the same year ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> vs. <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> ) is a measure of mobility based on dependent coding, while the other two comparisons ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> vs. <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> vs. <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mi>t</mi></msub></math> </ephtml> ), which are asked in two adjacent years, are measures of occupational mobility based on independent coding. Let's denote the mobility estimate from the dependent coding as <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> and that from the independent coding as <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>I</mi></msup></math> </ephtml> . We have no information on which coders code which occupational variables. However, we can assume that dependent coding is similar to coding two responses by a single coder. In contrast, independent coding is similar to coding two responses by two different coders, raising concerns about both intercoder and response reliability.</p> <p>As mentioned earlier, depending on the starting year of the survey, about half of the respondents from year <emph>t</emph> can be linked backwardly, and the other half can be linked forwardly. As shown in Figure 1, from the backwardly linked sample, two occupation information of <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and one occupation information of year <emph>t</emph> are collected: (<reflink idref="bib1" id="ref82">1</reflink>) <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> , (<reflink idref="bib2" id="ref83">2</reflink>) <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> , and (<reflink idref="bib3" id="ref84">3</reflink>) <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> . From the forwardly linked sample, one occupation information of <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and two occupation information of year <emph>t</emph> are collected: (<reflink idref="bib2" id="ref85">2</reflink>) <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> , (<reflink idref="bib3" id="ref86">3</reflink>) <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> , and (<reflink idref="bib4" id="ref87">4</reflink>) <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mi>t</mi></msub></math> </ephtml> . With this information, we can calculate occupational mobility between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and <emph>t</emph> in four different ways. Using the backwardly linked sample, two occupational mobilities are estimated: one is based on the change between a previous occupation and a current occupation, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> (let's call it <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> ) and the other is based on the change between two current occupations, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> ). Likewise, using the forwardly linked sample, two mobilities can be estimated: one is based on the change between a previous occupation and a current occupation, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><msup><mi>D</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup></mrow></msup></math> </ephtml> ), and the other is based on the change between two previous occupations, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mi>t</mi></msub></math> </ephtml> ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> ). Of the four measures, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><msup><mi>D</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup></mrow></msup></math> </ephtml> are conceptually identical. The only difference is the sample: one is a backwardly linked sample and the other is a forwardly linked sample.</p> <p>Graph: Figure 1. Four measures of occupational mobility between adjacent years using the Current Population Survey (CPS) mini-panel. Note: C refers to the current occupation and P refers to the main occupation in the previous year.</p> <hd id="AN0192656321-7">Sources of the Discrepancies Across Four Measures of Occupational Mobility</hd> <p>Although the four mobility measures discussed above measure slightly different occupational mobility, they are expected to yield similar outcomes in the absence of measurement problems. There are four potential sources of discrepancy between the four measures of occupational mobility. The first source is conceptual differences in occupation. The main occupation of the previous year ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> ) can be different from the occupation surveyed in March of that year ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> ). The discrepancy between mobility using two current occupations, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> , and that using two previous occupations, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> , can quantify the impact of the conceptual difference in measuring occupational mobility. We expect this difference to be small.</p> <p>Second, the respondent's recall error of the previous occupation can cause the discrepancy ([<reflink idref="bib35" id="ref88">35</reflink>]; [<reflink idref="bib48" id="ref89">48</reflink>]; [<reflink idref="bib45" id="ref90">45</reflink>]). In the backwardly linked sample, while <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> is the correct response to the main occupation of <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> , the respondent can erroneously answer <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> , which is the same or similar to <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> but different from <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> . This is especially likely when the respondent changed occupation during the second half of year <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and worked as <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> more than six months before March of year <emph>t</emph>. However, the magnitude of recall errors is presumably small. Furthermore, assuming the likelihood of a respondent's recall error is stochastically constant over time, the mismatch due to recall errors would not increase or decrease longitudinally. Regarding the difference between current and previous occupations, because the previous occupation is more prone to recall error than the current occupation, the estimated mobility based on <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mi>t</mi></msub></math> </ephtml> ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> ) will be higher than that based on <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> ). There is no difference in the occupational coding methods between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> . Thus, the discrepancy between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> reflects how serious these recall errors are, at least partially.</p> <p>The third source is low response reliability. Descriptions of the same occupations by the same respondents can be slightly inconsistent in two different surveys. The accuracy and consistency of descriptions may vary by demographic characteristics. One possibility is that less-educated respondents might provide less precise descriptions so that their occupational mismatches would be higher than those of highly educated respondents. The other possibility is that occupational mismatches can be higher for complex occupations because it is difficult to accurately describe highly complicated tasks, and respondents can provide slightly different descriptions of their occupations ([<reflink idref="bib24" id="ref91">24</reflink>]). Yet, another possibility is that the duties and activities of the same job can change. The same occupation can evolve into different occupations. This is why occupational classification codes are updated. With the evolution of the New Economy, the tasks of the same jobs/occupations can shift differently depending on the industrial context of those works. In this case, we can expect coding mismatches to increase over time. Proxy interviews can be another source of low response reliability ([<reflink idref="bib26" id="ref92">26</reflink>]; [<reflink idref="bib60" id="ref93">60</reflink>]). If one interview between two adjacent years is conducted by a proxy respondent while the other is conducted with the self respondent, their descriptions can differ. If the proportion of proxy interviews increases over time, the mismatch between the two occupational measures, as well as the discrepancies across the four occupational mobility measures, could increase.</p> <p>The fourth source is low intercoder reliability. Even though the respondent offered the same or similar descriptions for both current and previous occupations, different coders can assign different occupation codes. Longitudinally, assuming that intercoder reliability is stable over time, the fourth source does not necessarily cause an increase or decrease in mismatches. However, mismatch rates can change if there is less or more frequent turnover of coders and/or if there are other sources of low intercoder reliability, such as the evolution of new occupations.</p> <p>Among the four sources, the first and second sources are irrelevant to coding mismatches between coders, while the third and fourth sources are relevant. The discrepancy between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> quantifies the severity of the first two sources, while the discrepancies between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><msup><mi>D</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup></mrow></msup></math> </ephtml> quantify the severity of the third and fourth sources. In the presence of the third and fourth sources, mobility measures based on independent coding, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> , would be overestimated, while those based on dependent coding, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> (and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><msup><mi>D</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup></mrow></msup></math> </ephtml> ) will be closer to the true estimate. In this study, we aim to investigate how serious the discrepancy between mobilities based on dependently and independently coded occupations is, whether coding mismatches have increased in recent decades, and how this affects estimates of occupational career mobility.</p> <hd id="AN0192656321-8">Analytic Strategy</hd> <p></p> <hd id="AN0192656321-9">Data</hd> <p>We constructed mini-panels using the CPS-ASEC in March between 1991 and 2020 from the Integrated Public Use Microdata Series ([<reflink idref="bib17" id="ref94">17</reflink>]). While using all the basic monthly CPS will provide more cases, we use the March sample because the quesiton about the last year's main occupation is only available in the CPS-ASEC. Utilizing the rotation panel aspect, the CPS-ASEC from 1991 to 2020 can turn into 29 mini-panels from 1991–1992 to 2019–2020. During this period, the census occupational classification coding scheme changed twice in 2003 (from the 1990 classification to the 2000 classification) and 2011 (from the 2000 classification to the 2010 classification). While the transition from the 2000 classification to the 2010 classification introduced minimal noise, the transition from the 1990 classification to the 2000 classification created significant consistency problems in coding. Thus, we drop the mini-panels involving 2003 from the sample (i.e., the backwardly linked 2002–2003 sample and the forwardly linked 2003–2004 sample). We also drop mini-panels of 1994–1995 and 1995–1996 from the analyses because longitudinal weight is not available for 1995. With these restrictions, we are left with 25 mini-panels ranging from 1991–1992 to 2019–2020.</p> <p>As for sample restrictions, only those who appear in two adjacent years are selected. Those who appear in only one year are dropped from the analyses.[<reflink idref="bib12" id="ref95">12</reflink>] A backwardly linked sample utilizes mini-panels from 1991–1992 to 2018–2019, and a forwardly linked sample consists of mini-panels from 1992–1993 to 2019–2020. There are 24 mini-panels for both backwardly and forwardly linked samples. The total number of linked individuals is 369,016 for the backwardly linked sample and 362,245 for the forwardly linked sample. The observed respondent characteristics of the backwardly and forwardly linked samples are almost identical (see Supplemental Appendix Table A1).</p> <p>There are several additional sample restrictions. The analytic sample is limited to the 25–64 years old working-age population. If the age gap between two linked surveys exceeds two years, those cases are dropped. When gender and/or race/ethnicity do not match in the linked data, we drop them as well. We further exclude those in the military, unpaid family workers, and the self-employed in either of the two linked surveys. Individuals whose occupations are imputed are removed from the analyses as imputation affects mobility estimates ([<reflink idref="bib34" id="ref96">34</reflink>]). Individuals who are not at work at the first survey are also dropped. As for the occupational classification scheme, we use the IPUMS 1990 occupational classification for all years. We check the robustness of our findings with the other classification systems. Both three-digit detailed occupations and one-digit broad occupations are analyzed.</p> <hd id="AN0192656321-10">Statistical Approach</hd> <p>Using these datasets, we conduct three sets of analyses. First, we explore the changes in occupations between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and <emph>t</emph> descriptively. Second, we estimate the linear probability models (LPMs) of occupational mobility and then compare the predicted probabilities of occupational mobility across four measures— <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> , <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><msup><mi>D</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup></mrow></msup></math> </ephtml> , <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> , and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> . Equation (<reflink idref="bib1" id="ref97">1</reflink>) shows the model estimating <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> , the mobility between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> . <ephtml> <math display="block" xmlns="http://www.w3.org/1998/Math/MathML"><mi>D</mi><mo stretchy="false">(</mo><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>≠</mo><msub><mi>C</mi><mi>t</mi></msub><msub><mo stretchy="false">)</mo><mi>i</mi></msub><mo>=</mo><mi>α</mi><mo>+</mo><mo>∑</mo><msub><mi>β</mi><mi>j</mi></msub><msub><mi>X</mi><mrow><mi>i</mi><mi>j</mi></mrow></msub><mo>+</mo><msup><mi>R</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup><mi mathvariant="normal">Γ</mi><mo>+</mo><msup><mi>S</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup><mi mathvariant="normal">Δ</mi><mo>+</mo><msup><mi>O</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup><mi mathvariant="normal">Λ</mi><mo>+</mo><msub><mi>T</mi><mi>t</mi></msub><mo>+</mo><msub><mi>ε</mi><mi>i</mi></msub><mo>,</mo></math> </ephtml></p> <p>Graph</p> <p>where <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>D</mi><mo stretchy="false">(</mo><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>≠</mo><msub><mi>C</mi><mi>t</mi></msub><msub><mo stretchy="false">)</mo><mi>i</mi></msub></math> </ephtml> refers to a dummy variable indicating whether two current occupations between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and <emph>t</emph> are the same (0) or not (<reflink idref="bib1" id="ref98">1</reflink>). <emph>X</emph> is a set of demographic variables including gender, age groups (25–34, 35–44, and 45–54), race/ethnicity (whites, blacks, Hispanics, Asian Americans, and others), levels of education (LTHS, HSG, SC, and BA+), marital status (single, currently married, and separated/divorced/widowed), metro status, and nine census regions. We use all covariates reported in the first-year survey of the two adjacent years. Using information from the second-year survey does not alter the results. <emph>R</emph> is a set of proxy interview indicators distinguishing between never, only in the first survey, only in the second survey, and both times. <emph>S</emph> is a set of imputed work hours indicators for imputed weeks worked or usual hours worked per week variables. <emph>O</emph> is a set of one-digit occupation dummy variables. We start without controlling for <emph>O</emph> in the model and then add <emph>O</emph>. In some models, we test with three-digit occupations instead of one-digit ones. <emph>T</emph> is a set of survey year dummies. The same models are estimated for all four mobility measures.</p> <p>Assuming that the estimates of occupational mobility of dependent coding are true, the mismatch between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> when there is no change between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> (= <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>P</mi><mi>r</mi><mo stretchy="false">[</mo><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>≠</mo><msub><mi>C</mi><mi>t</mi></msub><mo fence="false" stretchy="false">|</mo><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>C</mi><mi>t</mi></msub><mo stretchy="false">]</mo></math> </ephtml> ) indicates the probability of false mobility between two independently coded occupations, while match of all three occupational codes (= <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>P</mi><mi>r</mi><mo stretchy="false">[</mo><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>C</mi><mi>t</mi></msub><mo fence="false" stretchy="false">|</mo><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>C</mi><mi>t</mi></msub><mo stretchy="false">]</mo></math> </ephtml> ) quantifies the probability of correct immobility. Likewise, the mismatch between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> when there is a change from <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> to <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> (= <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>P</mi><mi>r</mi><mo stretchy="false">[</mo><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>≠</mo><msub><mi>C</mi><mi>t</mi></msub><mo fence="false" stretchy="false">|</mo><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>≠</mo><msub><mi>C</mi><mi>t</mi></msub><mo stretchy="false">]</mo></math> </ephtml> ) implies the probability of correct mobility and the match between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> (= <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>P</mi><mi>r</mi><mo stretchy="false">[</mo><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>C</mi><mi>t</mi></msub><mo fence="false" stretchy="false">|</mo><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>≠</mo><msub><mi>C</mi><mi>t</mi></msub><mo stretchy="false">]</mo></math> </ephtml> ) implies the probability of false immobility. The same logic can be applied to the forwardly linked sample. Based on this rationale, as a third analysis, we limit our sample to those whose occupations are not changed with the dependent coding (i.e., <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> ) and then estimate the LPM of the mismatch between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> (and the LPM of the mismatch between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mi>t</mi></msub></math> </ephtml> for the forwardly linked sample). With this analysis, we explore whether coding mismatches are associated with false mobility.</p> <p>The statistical model for the third analysis is basically the same as equation (<reflink idref="bib1" id="ref99">1</reflink>) except that <emph>T</emph> is now a continuous variable. With a continuous time variable, we test whether the mismatch increases, net of other controls. Additional models, which include the interaction terms between <emph>T</emph> and other covariates, are estimated as well to check whether the time trend varies depending on covariates and occupations. To assess the extent to which the change in the mismatch is associated with the changes in the composition of covariates and the unexplained coefficient effect, we conduct the Oaxaca-Blinder decomposition by dividing our data into the 1992–1996 panel and the 2015–2019 panel.</p> <hd id="AN0192656321-11">Results</hd> <p></p> <hd id="AN0192656321-12">Descriptive Analyses</hd> <p>Figure 2 shows the changes in occupations between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and <emph>t</emph> across four measures. The changes measured with two independently coded occupations ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> ) are astonishingly higher than those measured with two dependently coded occupations ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><msup><mi>D</mi><mi mathvariant="normal">′</mi></msup></mrow></msup></math> </ephtml> ). Throughout the whole period, the gaps are consistently large. Using the broad one-digit occupation does not fix the problem. The average rates of three-digit and one-digit occupation changes are 43.8 percent and 27.6 percent for <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> , while for <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> , they are 7.1 percent and 3.0 percent, respectively. The rates of three-digit occupational changes are more than six times larger for <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> than those for <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> . For a one-digit occupation, the occupational change rate of <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> is about nine times larger than that of <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> . Another noticeable discrepancy between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> is that the former shows the upward trend of occupational mobility over time, while the latter does not show such a trend clearly.</p> <p>Graph: Figure 2. Changes in occupational status between two adjacent years: (a) backwardly linked sample: t−1 year rotation group and (b) forwardly linked sample: t year rotation group.</p> <p>These discrepancies are unlikely to be associated with the conceptual differences between the current occupation and the previous main occupation of the last year. When the outcomes of the backwardly linked sample (Figure 2[I]) are compared to those of the forwardly linked sample (Figure 2[II]), the results of <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> are very similar to those of <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> , and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> are very similar to <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><msup><mi>D</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup></mrow></msup></math> </ephtml> . Regardless of how occupations are measured, occupational mobility turns out to be substantially higher when two independently coded occupations are used than when two dependently coded occupations are used. This does not mean that there is no difference between the mobility of current occupations ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> ) and that of previous main occupations ( <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> ). As shown in Table 1, the mobility of the two previous occupations is higher than that of the two current occupations. Previous occupation is more prone to recall errors, and thus, as expected, <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> is higher than <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> . However, the discrepancies between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> are small, at best. Therefore, conceptual differences in occupation and recall errors are not major sources of the discrepancies.</p> <p>Table 1. Three-Digit Occupational Mobility Between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and t.</p> <p>Graph</p> <p> <ephtml> <table><colgroup><col align="left" /><col align="center" /><col align="center" /><col align="center" /><col align="center" /><col align="center" /><col align="center" /><col align="center" /><col align="center" /></colgroup><thead><tr><th align="left" /><th align="center" colspan="4">Proportion of Occupation Changes</th><th align="center" colspan="4">Gap Between Measures</th></tr><tr><th align="left" /><th align="center">(a)</th><th align="center">(b)</th><th align="center">(c)</th><th align="center">(d)</th><th align="center">(b)–(a)</th><th align="center">(d)–(c)</th><th align="center">(b)–(d)</th><th align="center">(a)–(c)</th></tr><tr><th align="left" /><th align="center"><p><math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup xmlns=""><mi>m</mi><mi>D</mi></msup></math></p></th><th align="center"><p><math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup xmlns=""><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math></p></th><th align="center"><p><math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup xmlns=""><mi>m</mi><mrow><msup><mi>D</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup></mrow></msup></math></p></th><th align="center"><p><math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup xmlns=""><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math></p></th><th align="left" /><th align="left" /><th align="left" /><th align="left" /></tr></thead><tbody><tr><td>Total</td><td>0.071</td><td>0.438</td><td>0.087</td><td>0.474</td><td>0.367</td><td>0.387</td><td>−0.036</td><td>−0.016</td></tr><tr><td>By one-digit occupation</td><td /><td /><td /><td /><td /><td /><td /><td /></tr><tr><td>– Executive/administrative/managerial</td><td>0.056</td><td>0.498</td><td>0.067</td><td>0.518</td><td>0.442</td><td>0.451</td><td>−0.019</td><td>−0.011</td></tr><tr><td>– Management related</td><td>0.073</td><td>0.496</td><td>0.084</td><td>0.521</td><td>0.423</td><td>0.437</td><td>−0.025</td><td>−0.011</td></tr><tr><td>– Professional specialty</td><td>0.052</td><td>0.338</td><td>0.068</td><td>0.375</td><td>0.286</td><td>0.306</td><td>−0.037</td><td>−0.016</td></tr><tr><td>– Technicians and related support</td><td>0.075</td><td>0.420</td><td>0.080</td><td>0.463</td><td>0.346</td><td>0.383</td><td>−0.042</td><td>−0.005</td></tr><tr><td>– Sales</td><td>0.095</td><td>0.443</td><td>0.110</td><td>0.474</td><td>0.347</td><td>0.364</td><td>−0.031</td><td>−0.015</td></tr><tr><td>– Administrative support</td><td>0.081</td><td>0.539</td><td>0.098</td><td>0.575</td><td>0.459</td><td>0.476</td><td>−0.035</td><td>−0.018</td></tr><tr><td>– Service</td><td>0.076</td><td>0.353</td><td>0.102</td><td>0.407</td><td>0.277</td><td>0.305</td><td>−0.054</td><td>−0.026</td></tr><tr><td>– Operator/fabricator/laborer</td><td>0.082</td><td>0.471</td><td>0.100</td><td>0.510</td><td>0.390</td><td>0.410</td><td>−0.039</td><td>−0.018</td></tr><tr><td>– Other</td><td>0.071</td><td>0.455</td><td>0.083</td><td>0.492</td><td>0.384</td><td>0.409</td><td>−0.037</td><td>−0.012</td></tr></tbody></table> </ephtml> </p> <p>1 <emph>Note:</emph> Samples are limited to continuously employed individuals. <emph>C</emph> refers to the current occupation, and <emph>P</emph> refers to the main occupation in the previous year. "Other" category ofone-digit occupation includes farming/forestry/fishing/production/craft/repair. Longitudinal weights are applied.</p> <p>In Table 1, we also explore whether the large discrepancies between mobility measures are consistently observed across broad one-digit occupational categories. When <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> are compared (i.e., (b)–(a) column in Table 1), the discrepancy is relatively large for management and administrative-related occupations and relatively small for service or sales occupations. Unlike other white-collar occupations, professional occupations exhibit relatively small discrepancies. Importantly, for all one-digit occupations, mobility rates are low when dependently coded occupations are compared, but they are high when independently coded occupations are compared. These results imply that the large discrepancies between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> (and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><msup><mi>D</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup></mrow></msup></math> </ephtml> ) are caused by the mismatches between coders, which are associated with either low response reliability or low intercoder reliability, or both.</p> <p>Table 2 presents the three-digit occupational match rates between two occupational measures in the same year (e.g., current occupation in year <emph>t</emph> and previous occupation asked in year <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>+</mo><mn>1</mn></math> </ephtml> ). The two occupational measures differ conceptually, but they are expected to be similar, and thus, absent measurement error, the match rate should be high. The match rates between current and previous occupations are only around 55 percent in both panels. The match rates are highest for professional specialty occupations and lowest for administrative support occupations. These results are consistent with the previous study, which reported that professional occupations are less prone to coding mismatches ([<reflink idref="bib38" id="ref100">38</reflink>]).</p> <p>Table 2. Match Rate Between Three-Digit Current Occupation and Previous Occupation.</p> <p>Graph</p> <p> <ephtml> <table><colgroup><col align="left" /><col align="center" /><col align="center" /><col align="center" /><col align="center" /></colgroup><thead><tr><th align="left" /><th align="center" colspan="2"><p><math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub xmlns=""><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math></p> versus <p><math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub xmlns=""><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math></p></th><th align="center" colspan="2"><p><math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub xmlns=""><mi>C</mi><mi>t</mi></msub></math></p> versus <p><math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub xmlns=""><mi>P</mi><mi>t</mi></msub></math></p></th></tr><tr><th align="left" /><th align="center" colspan="2">Matched</th><th align="center" colspan="2">Matched</th></tr><tr><th align="left" /><th align="center">O</th><th align="center">X</th><th align="center">O</th><th align="center">X</th></tr></thead><tbody><tr><td>Total</td><td>0.555</td><td>0.445</td><td>0.553</td><td>0.447</td></tr><tr><td>By one-digit occupation</td><td /><td /><td /><td /></tr><tr><td>– Executive/administrative/managerial</td><td>0.499</td><td>0.501</td><td>0.499</td><td>0.501</td></tr><tr><td>– Management related</td><td>0.499</td><td>0.501</td><td>0.499</td><td>0.501</td></tr><tr><td>– Professional specialty</td><td>0.654</td><td>0.346</td><td>0.651</td><td>0.349</td></tr><tr><td>– Technicians and related support</td><td>0.565</td><td>0.435</td><td>0.564</td><td>0.436</td></tr><tr><td>– Sales</td><td>0.548</td><td>0.452</td><td>0.548</td><td>0.452</td></tr><tr><td>– Administrative support</td><td>0.460</td><td>0.540</td><td>0.453</td><td>0.547</td></tr><tr><td>– Service</td><td>0.639</td><td>0.361</td><td>0.635</td><td>0.365</td></tr><tr><td>– Operator/fabricator/laborer</td><td>0.526</td><td>0.474</td><td>0.521</td><td>0.479</td></tr><tr><td>– Other</td><td>0.535</td><td>0.465</td><td>0.532</td><td>0.468</td></tr></tbody></table> </ephtml> </p> <p>2 <emph>Note:</emph> Samples are limited to continuously working individuals. <emph>C</emph> refers to the current occupation, and <emph>P</emph> refers to the main occupation in the previous year. "Other" category ofone-digit occupation includes farming/forestry/fishing/production/craft/repair. Longitudinal weights are applied.</p> <p>Assuming that mobility measures of <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><msup><mi>D</mi><mi mathvariant="normal">′</mi></msup></mrow></msup></math> </ephtml> , which are based on dependently coded occupations, are true, Figure 3 depicts false and correct mobilities and immobilities. For example, if <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> indicates change and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> also indicates change, it is a correct mobility. If <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> indicates no change while <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> indicates change, it is false immobility. On the other hand, if <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> indicates no change, changes in <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> indicate false mobility, and no changes in <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> indicate correct immobility. Regardless of the digits of the occupation codes and the ways of linking two adjacent years, Figure 3 shows the rise in false mobility over the last three decades. This result implies that the reliability of occupational coding may have eroded.</p> <p>Graph: Figure 3. Mobility estimates, assuming that the dependent coding is correct: (a) backwardly linked sample: t−1 year rotation group and (b) forwardly linked sample: t year rotation group.</p> <hd id="AN0192656321-13">Divergence in the Estimates of Occupational Mobility Across Measures</hd> <p>To check whether the further divergence between mobility based on the independently coded occupations and that based on the dependently coded occupations is evident net of demographic and survey-related technical covariates, we estimate equation (<reflink idref="bib1" id="ref101">1</reflink>) and calculate the predicted occupational mobility. Figure 4 illustrates the changes. The left-side graphs show the predicted occupational mobility in both one-digit and three-digit occupations, and the right-side graphs show the evolvement of the discrepancies in mobility between independently and dependently coded occupations over time, setting the gap in 1992 as zero.</p> <p>Graph: Figure 4. Predicted intragenerational occupational mobility: (a) backwardly linked sample: t−1 year rotation group and (b) forwardly linked sample: t year rotation group. Note: Sample is limited to continuously working individuals. Control variables include proxy interview indicators, imputed work hours variables, gender, age, race/ethnicity, levels of education, marital status, metro status, and census region. Plot (ii) sets the discrepancy between measures of 1992 at zero. Longitudinal weights are applied.</p> <p>Figure 4 once again shows the strikingly large discrepancy between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> (and between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><msup><mi>D</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup></mrow></msup></math> </ephtml> ). Regarding the rising trend in occupational mobility, all measures using three-digit occupations show a clear upward swing, but some measures using one-digit occupations do not show a clear trend. However, when we estimate the slopes of the time variable (check Supplemental Appendix Table A2 for details), all four measures show a statistically significantly positive coefficient, supporting the rise in intragenerational occupational mobility ([<reflink idref="bib30" id="ref102">30</reflink>]). Notably, the growth in occupational mobility with independent coding is three to four times steeper than that with dependent coding.</p> <p>When the discrepancy in 1992 is set to zero, the right-side graphs of Figure 4 demonstrate increasing discrepancies between the independent and dependent codings regardless of the digits of occupation. Over the study period, the discrepancy between the two mobility measures grew by 2.4 to 8.2 percentage points. Overall, our findings corroborate that there is a significant discrepancy in occupational mobility between independently coded and dependently coded occupations and that this discrepancy has widened in recent decades.</p> <hd id="AN0192656321-14">Growth in Coding Mismatches Between Coders</hd> <p>The likely cause of the discrepancy demonstrated above is coding mismatches between coders. Both low response reliability and low intercoder reliability can lead to coding mismatches. Response reliability is associated with respondent demographics and proxy interviews, while intercoder reliability is unaffected by these covariates. Due to data limitations, it is not possible to precisely assess the extent to which coding mismatches between coders are associated with response reliability and/or intercoder reliability. However, by estimating the associations between demographic and survey-method-related covariates and coding mismatches and checking the degree to which these covariates account for the coding mismatches, we can infer whether both response reliability and intercoder reliability are equally important or one is relatively more important than the other. If changes in response reliability are the main driver of the growing discrepancy, these covariates will account for at least a significant portion of the growth in coding mismatches.</p> <p>As shown in Figure 3, false mobility is the main driver of the rise in the discrepancy across mobility measures. Thus, we estimate the LPMs of the coding mismatch between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> and <emph>t</emph> after limiting our sample to those who did not change occupations when measured in the dependent coding (i.e., <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> ). For non-changers, all three occupations, which are <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>,</mo><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> , and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> in the backwardly linked sample, and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> , <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> , and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mi>t</mi></msub></math> </ephtml> in the forwardly linked sample, are supposed to match. Non-matches are most likely the result of disagreement between coders, although the possibility of real differences cannot be ruled out. Our main interest is whether the match rates (i.e., <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>P</mi><mi>r</mi><mo stretchy="false">(</mo><msub><mi>C</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>C</mi><mi>t</mi></msub><mo fence="false" stretchy="false">|</mo><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> ) and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>P</mi><mi>r</mi><mo stretchy="false">(</mo><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>P</mi><mi>t</mi></msub><mo fence="false" stretchy="false">|</mo><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub><mo>=</mo><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> )) increase over time and, if so, whether the covariates controlled for in the models can account for the increase. Because the trends of false mobility are similar between the forwardly and backwardly linked samples, we combine them in our regression analyses. The estimates segregated by the link method yield basically the same results.</p> <p>Descriptively, the mismatch rate has increased from 39.2 percent in 1992 to 45.8 percent in 2019. As shown in Model 1 of Table 3, the mismatch rate has increased by 0.28 percent per year. The control of demographic and survey-method-related covariates in Model 2 does not account for the rise in mismatch rates. Instead, the coefficient of time in Model 2 increases slightly to 0.31 percent per year. This is not because the controlled covariates are irrelevant to mismatches. As anticipated, mismatch rates are higher among proxy interviews than among self-responses. The mismatch rate of the previous occupations in the past is 1.3 percentage points higher than that of the current occupations. The imputation of weeks worked is also associated with a 7.6 percentage point increase in the mismatch rate. All demographic covariates, such as gender, race, education, marital status, and census regions, are relevant to mismatches. Nevertheless, they do not account for the growth in mismatches.[<reflink idref="bib13" id="ref103">13</reflink>] Model 2 suggests that the distributional changes in neither demographic nor survey-method-related covariates can explain the growth in coding mismatches. From this, we can infer that changes in response reliability are an unlikely cause of the growing discrepancy.</p> <p>Table 3. LPM Estimates of the Mismatch Between Independently Coded Occupations in Two Adjacent Years Among Those Who Did Not Change Three-Digit Occupations in Dependent Coding.</p> <p>Graph</p> <p> <ephtml> <table><colgroup><col align="left" /><col align="center" /><col align="center" /><col align="char" char="." /><col align="center" /></colgroup><thead><tr><th align="left" /><th align="center" colspan="2">Model 1</th><th align="center" colspan="2">Model 2</th></tr><tr><th align="left" /><th align="center">Coef.</th><th align="center">SE</th><th align="center">Coef.</th><th align="center">SE</th></tr></thead><tbody><tr><td>Time (1992 = 0)</td><td>0.003***</td><td>(0.000)</td><td>0.003***</td><td>(0.000)</td></tr><tr><td>Forwardly linked sample</td><td>0.013***</td><td>(0.001)</td><td>0.014***</td><td>(0.001)</td></tr><tr><td>Proxy interview (ref. none)</td><td /><td /><td /><td /></tr><tr><td>– First survey</td><td /><td /><td>0.039***</td><td>(0.002)</td></tr><tr><td>– Second survey</td><td /><td /><td>0.039***</td><td>(0.002)</td></tr><tr><td>– Both surveys</td><td /><td /><td>0.019***</td><td>(0.001)</td></tr><tr><td>Imputed work hours</td><td /><td /><td /><td /></tr><tr><td>– Weekly work hours</td><td /><td /><td>0.008</td><td>(0.007)</td></tr><tr><td>– Weeks worked</td><td /><td /><td>0.076***</td><td>(0.008)</td></tr><tr><td>Women</td><td /><td /><td>−0.006***</td><td>(0.001)</td></tr><tr><td>Age (ref. 25–34)</td><td /><td /><td /><td /></tr><tr><td>– 35–44</td><td /><td /><td>−0.016***</td><td>(0.002)</td></tr><tr><td>– 45–54</td><td /><td /><td>−0.025***</td><td>(0.002)</td></tr><tr><td>– 55–64</td><td /><td /><td>−0.031***</td><td>(0.002)</td></tr><tr><td>Race/Ethnicity (ref. White)</td><td /><td /><td /><td /></tr><tr><td>– Black</td><td /><td /><td>0.034***</td><td>(0.002)</td></tr><tr><td>– Asian</td><td /><td /><td>0.010∗∗</td><td>(0.003)</td></tr><tr><td>– Hispanic</td><td /><td /><td>0.022***</td><td>(0.002)</td></tr><tr><td>– Other</td><td /><td /><td>0.020***</td><td>(0.006)</td></tr><tr><td>Education (ref. LTHS)</td><td /><td /><td /><td /></tr><tr><td>– HSG</td><td /><td /><td>0.032***</td><td>(0.003)</td></tr><tr><td>– SC</td><td /><td /><td>0.027***</td><td>(0.003)</td></tr><tr><td>– BA+</td><td /><td /><td>−0.019***</td><td>(0.003)</td></tr><tr><td>Marital status (ref. never married single)</td><td /><td /><td /><td /></tr><tr><td>– Currently married</td><td /><td /><td>−0.030***</td><td>(0.002)</td></tr><tr><td>– Separated/divorced/widowed</td><td /><td /><td>−0.015***</td><td>(0.002)</td></tr><tr><td>Metro status</td><td /><td /><td>0.018***</td><td>(0.002)</td></tr><tr><td>Census region division (ref. New England)</td><td /><td /><td /><td /></tr><tr><td>– Middle Atlantic</td><td /><td /><td>−0.000</td><td>(0.003)</td></tr><tr><td>– East North Central</td><td /><td /><td>0.006</td><td>(0.003)</td></tr><tr><td>– West North Central</td><td /><td /><td>−0.000</td><td>(0.003)</td></tr><tr><td>– South Atlantic</td><td /><td /><td>0.013***</td><td>(0.003)</td></tr><tr><td>– East South Central</td><td /><td /><td>0.009∗</td><td>(0.004)</td></tr><tr><td>– West South Central</td><td /><td /><td>0.022***</td><td>(0.003)</td></tr><tr><td>– Mountain</td><td /><td /><td>0.025***</td><td>(0.004)</td></tr><tr><td>– Pacific</td><td /><td /><td>0.005</td><td>(0.003)</td></tr><tr><td>Constant</td><td>0.378***</td><td>(0.001)</td><td>0.357***</td><td>(0.004)</td></tr><tr><td>Observations</td><td align="center" colspan="2">628,269</td><td align="center" colspan="2">628,269</td></tr><tr><td><italic>R</italic>2</td><td align="center" colspan="2">0.002</td><td align="center" colspan="2">0.009</td></tr></tbody></table> </ephtml> </p> <ulist> <item>3 <emph>Note:</emph> Samples are limited to continuously working individuals whose <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> are the same. Robust standard errors are in parentheses. Longitudinal weights are applied.</item> <item>4 <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mrow /><mo>*</mo></msup><mi>p</mi><mo><</mo><mn>.05</mn></math> </ephtml> , <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mrow /><mrow><mo>*</mo><mo>*</mo></mrow></msup><mi>p</mi><mo><</mo><mn>.01</mn></math> </ephtml> , <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mrow /><mrow><mo>*</mo><mo>*</mo><mo>*</mo></mrow></msup><mi>p</mi><mo><</mo><mn>.001</mn></math> </ephtml> (two-tailed tests).</item> </ulist> <p>Another noteworthy point of Table 3 is the very low <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>R</mi><mn>2</mn></msup></math> </ephtml> . A high <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>R</mi><mn>2</mn></msup></math> </ephtml> is not expected in LPMs, but the low <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>R</mi><mn>2</mn></msup></math> </ephtml> indicates that the controlled covariates can hardly explain not just the increase in coding mismatches but coding mismatches in general. Coding mismatches seem to be stochastic procedures mostly independent of the demographic and survey-method-related covariates.</p> <p>Next, we explore whether the change in mismatches varies across occupations. Table 4 shows the result with one-digit occupation dummies and their interactions with time added on top of the control variables of Model 2. Consistent with the descriptive statistics of Table 1, coding mismatches are higher for management and administrative support occupations than other occupations, while they are lower for professional, technician, and service-related occupations. Regarding the longitudinal changes, sales and administrative support occupations exhibit faster rises than others, while professional occupations show relatively slower rises. It is noteworthy that two prestigious occupations—executive/administrative/managerial and professional specialty occupations—belong to the groups of the highest and lowest match rates, respectively. Over time, the gap in mismatches between these two occupations has widened further. Importantly, none of the negative interaction effects of time and occupation are large enough to nullify the positive coefficient of the main effect of time. All one-digit occupations show a growing trend in mismatches.</p> <p>Table 4. Linear Probability Model (LPM) Estimates of the Mismatch Between Independently Coded Occupations in Two Adjacent Years Among Those Who Did Not Change Three-Digit Occupations in Dependent Coding, the Variation Across One-Digit Occupations.</p> <p>Graph</p> <p> <ephtml> <table><colgroup><col align="left" /><col align="char" char="." /><col align="char" char="." /></colgroup><thead><tr><th align="left" /><th align="center" colspan="2">Model 3</th></tr><tr><th align="left" /><th align="center">Coef.</th><th align="center">SE</th></tr></thead><tbody><tr><td>Time (1992 = 0)</td><td>0.004***</td><td>(0.000)</td></tr><tr><td>Forwardly linked sample</td><td>0.014***</td><td>(0.001)</td></tr><tr><td>One-digit occupation</td><td /><td /></tr><tr><td>(ref. executive/administrative/managerial)</td><td /><td /></tr><tr><td>– Management related</td><td>0.017∗</td><td>(0.007)</td></tr><tr><td>– Professional specialty</td><td>−0.135***</td><td>(0.005)</td></tr><tr><td>– Technicians and related support</td><td>−0.064***</td><td>(0.007)</td></tr><tr><td>– Sales</td><td>−0.116***</td><td>(0.006)</td></tr><tr><td>– Administrative support</td><td>0.011∗</td><td>(0.005)</td></tr><tr><td>– Service</td><td>−0.153***</td><td>(0.005)</td></tr><tr><td>– Operator/fabricator/laborer</td><td>−0.050***</td><td>(0.005)</td></tr><tr><td>– Other</td><td>−0.052***</td><td>(0.005)</td></tr><tr><td>Time × one-digit occupation</td><td /><td /></tr><tr><td>– Management related</td><td>−0.002***</td><td>(0.000)</td></tr><tr><td>– Professional specialty</td><td>−0.002***</td><td>(0.000)</td></tr><tr><td>– Technicians and related support</td><td>−0.001∗∗</td><td>(0.000)</td></tr><tr><td>– Sales</td><td>0.003***</td><td>(0.000)</td></tr><tr><td>– Administrative support</td><td>0.002***</td><td>(0.000)</td></tr><tr><td>– Service</td><td>−0.001***</td><td>(0.000)</td></tr><tr><td>– Operator/fabricator/laborer</td><td>−0.000</td><td>(0.000)</td></tr><tr><td>– Other</td><td>−0.001</td><td>(0.000)</td></tr><tr><td>Observations</td><td>628,269</td><td /></tr><tr><td><italic>R</italic>2</td><td>0.030</td><td /></tr></tbody></table> </ephtml> </p> <ulist> <item>5 <emph>Note:</emph> Samples are limited to continuously working individuals whose <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> are the same. "Other" category of one-digit occupation includesfarming/forestry/fishing/production/craft/repair. Control variables include proxy interview indicators, imputed work hours variables, gender, age, race/ethnicity, levels of education, marital status, metro status, and census region. Robust standard errors are in parentheses. Longitudinal weights are applied.</item> <item>6 <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mrow /><mo>*</mo></msup><mi>p</mi><mo><</mo><mn>.5</mn></math> </ephtml> , <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mrow /><mrow><mo>*</mo><mo>*</mo></mrow></msup><mi>p</mi><mo><</mo><mn>.01</mn></math> </ephtml> , <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mrow /><mrow><mo>*</mo><mo>*</mo><mo>*</mo></mrow></msup><mi>p</mi><mo><</mo><mn>.001</mn></math> </ephtml> (two-tailed tests).</item> </ulist> <p>To check the extent to which mismatch changes vary across three-digit occupations, we contrast the predicted mismatch rates net of demographic and survey-method-related covariates in 1992–1996 with those in 2015–2019, as shown in Figure 5. Here, we consider only the occupations with sample sizes of 100 or more in both periods. Occupations on the upper left are those that experienced an increase in mismatch, while occupations on the lower right are those that experienced a decrease. Those on the diagonal line are occupations with no change. In both periods, the match rates vary widely from around 10 percent to more than 80 percent. Despite this wide variation, most of the detailed occupations are on the upper left, while some professional specialty occupations, such as registered nurses, primary, and secondary teachers, and a small number of other occupations, are on the lower right side.</p> <p>Graph: Figure 5. Predicted mismatch rates of three-digit occupations in 1992–1996 and 2015–2019. Note: Limited to three-digit occupations with at least 100 cases in each time period. Symbol sizes are weighted by the total cases in each occupation. The predicted probabilities are based on the regression models that control for the same covariates of Model 2 of Table 3.</p> <hd id="AN0192656321-15">Oaxaca-Blinder Decomposition of the Changes Between 1992–1996 and 2015–2019</hd> <p>To assess how much rise in mismatches is associated with the changing compositions of respondent characteristics and their coefficient effects on mismatches, we conduct the Oaxaca-Blinder decomposition analysis by disaggregating our data into the 1992–1996 sample and the 2015–2019 sample. All covariates controlled for in Model 2 of Table 3, as well as one- and three-digit occupations, are added in the decomposition analyses. Table 5 presents the results.</p> <p>Table 5. Decomposition of the Change in the Mismatch Rates Between 1992–1996 and 2015–2019 Among Those Who Did Not Change Three-digit Occupations in Dependent Coding.</p> <p>Graph</p> <p> <ephtml> <table><colgroup><col align="left" /><col align="char" char="." /><col align="char" char="." /><col align="char" char="." /><col align="char" char="." /></colgroup><thead><tr><th align="left" /><th align="center" colspan="2">Model 1</th><th align="center" colspan="2">Model 2</th></tr><tr><th align="left" /><th align="center">Coef.</th><th align="center">SE</th><th align="center">Coef.</th><th align="center">SE</th></tr></thead><tbody><tr><td>Average</td><td /><td /><td /><td /></tr><tr><td>– 1992–1996</td><td>0.393***</td><td>(0.002)</td><td>0.393***</td><td>(0.002)</td></tr><tr><td>– 2015–2019</td><td>0.455***</td><td>(0.002)</td><td>0.455***</td><td>(0.002)</td></tr><tr><td>Total change</td><td>0.062***</td><td>(0.003)</td><td>0.062***</td><td>(0.003)</td></tr><tr><td>– Endowments</td><td>−0.016***</td><td>(0.002)</td><td>0.003</td><td>(0.003)</td></tr><tr><td>– Coefficients</td><td>0.077***</td><td>(0.003)</td><td>0.058***</td><td>(0.004)</td></tr><tr><td>Endowments</td><td /><td /><td /><td /></tr><tr><td>– Forwardly linked sample</td><td>−0.000</td><td>(0.000)</td><td>−0.000∗</td><td>(0.000)</td></tr><tr><td>– Proxy interview</td><td>−0.004***</td><td>(0.001)</td><td>−0.004***</td><td>(0.000)</td></tr><tr><td>– Imputed work hours</td><td>−0.000</td><td>(0.000)</td><td>−0.000</td><td>(0.000)</td></tr><tr><td>– Demographic</td><td>−0.003</td><td>(0.002)</td><td>−0.003</td><td>(0.002)</td></tr><tr><td>– One-digit occupation</td><td>−0.008***</td><td>(0.001)</td><td /><td /></tr><tr><td>– Three-digit occupation</td><td /><td /><td>0.010***</td><td>(0.003)</td></tr><tr><td>Coefficients</td><td /><td /><td /><td /></tr><tr><td>– Forwardly linked sample</td><td>−0.000∗∗</td><td>(0.000)</td><td>−0.000∗</td><td>(0.000)</td></tr><tr><td>– Proxy interview</td><td>−0.001∗∗</td><td>(0.000)</td><td>−0.001∗∗</td><td>(0.000)</td></tr><tr><td>– Imputed work hours</td><td>0.001***</td><td>(0.000)</td><td>0.001***</td><td>(0.000)</td></tr><tr><td>– Demographic</td><td>0.001</td><td>(0.001)</td><td>0.001</td><td>(0.001)</td></tr><tr><td>– One-digit occupation</td><td>−0.002***</td><td>(0.000)</td><td /><td /></tr><tr><td>– Three-digit occupation</td><td /><td /><td>−0.022***</td><td>(0.001)</td></tr><tr><td>– Constant</td><td>0.079***</td><td>(0.003)</td><td>0.080***</td><td>(0.003)</td></tr><tr><td>Observations</td><td>192,893</td><td /><td>192,893</td><td /></tr></tbody></table> </ephtml> </p> <ulist> <item>7 <emph>Note:</emph> Samples are limited to continuously working individuals whose <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>C</mi><mi>t</mi></msub></math> </ephtml> and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msub><mi>P</mi><mrow><mi>t</mi><mo>−</mo><mn>1</mn></mrow></msub></math> </ephtml> are the same. <emph>C</emph> refers to the current occupation, and <emph>P</emph> refers to the main occupation in the previous year. Demographic variables include gender, age, race and ethnicity, levels of education, marital status, metro status, and census region. To deal with the identification problem in detailed decomposition, all control variables are grand-mean centered ([<reflink idref="bib36" id="ref104">36</reflink>]). Robust standard errors are in parentheses. Longitudinal weights are applied.</item> <item>8 <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mrow /><mo>*</mo></msup><mi>p</mi><mo><</mo><mn>.05</mn></math> </ephtml> , <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mrow /><mrow><mo>*</mo><mo>*</mo></mrow></msup><mi>p</mi><mo><</mo><mn>.01</mn></math> </ephtml> , <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mrow /><mrow><mo>*</mo><mo>*</mo><mo>*</mo></mrow></msup><mi>p</mi><mo><</mo><mn>.001</mn></math> </ephtml> (two-tailed tests).</item> </ulist> <p>Between the two periods, the mismatch rate goes up by 6.2 percentage points. The decomposition results differ slightly depending on which digits of occupations are added to the models, but both models are consistent in exposing that compositional changes do not account for the rise in mismatches. The endowment effects are either negative in Model 1 or null in Model 2. The change in one-digit occupational structures seems to be associated with a small reduction in mismatches, while the change in three-digit occupational structures is associated with a small increase. In both cases, the extent to which changes in occupational structure account for the rise in mismatches is less than a 1 percentage point.</p> <p>The entire rise in mismatches is due to the coefficients effect. When the coefficient effect is decomposed into detailed variables, it turns out that the changes in the coefficients of proxy interviews and occupational structures are associated with small reductions in mismatches. That is, the rise in mismatches can be explained neither by the compositional changes nor the changes in the coefficients of individual covariates. The unexplained constant component accounts for a 7.9 to 8.0 percentage point rise in mismatches, more than the entire increase. This result suggests that the rise in mismatches is mostly a stochastic process that is independent of the controlled covariates, and is best explained by low intercoder reliability.</p> <hd id="AN0192656321-16">Do Continuous Measures Solve the Problem?</hd> <p>So far, our assessment is based on the discrete classification of occupations. Partly because of the reliability concerns with the discrete occupations, some scholars recommend utilizing gradational socioeconomic index (SEI) scores of occupations ([<reflink idref="bib27" id="ref105">27</reflink>]; [<reflink idref="bib38" id="ref106">38</reflink>]; [<reflink idref="bib59" id="ref107">59</reflink>]). Previous studies show that the latter measures are more reliable than the former ([<reflink idref="bib27" id="ref108">27</reflink>]; [<reflink idref="bib38" id="ref109">38</reflink>]; [<reflink idref="bib46" id="ref110">46</reflink>]; [<reflink idref="bib47" id="ref111">47</reflink>]). To explore whether using gradational SEIs reduces mismatches, we re-estimate mismatches after converting discrete occupations into the Duncan SEI (Duncan SEI) ([<reflink idref="bib13" id="ref112">13</reflink>]) and the occupation-based SEI with percentile ranks (occupational percentile rank) ([<reflink idref="bib59" id="ref113">59</reflink>]).[<reflink idref="bib14" id="ref114">14</reflink>]</p> <p>No clear definition or consensus exists on how much gap should be defined as occupational change when the gradational indices are used. We check the results with different thresholds: changes in scores larger than 1, 5, 10, 15, and 20. As shown in Supplemental Appendix Figures A.1 and A.2, no matter which thresholds we apply, using gradational SEI scores does not solve the mismatch issue. The estimates with gradational SEI scores also demonstrate the growing discrepancies across measures.</p> <p>These findings are not because our gradational measures are less reliable than those in the previous studies. The correlation coefficients of the gradational SEI scores between the two independently coded occupations range between 0.80 and 0.85. These numbers are equally high as in the previous studies ([<reflink idref="bib27" id="ref115">27</reflink>]; [<reflink idref="bib38" id="ref116">38</reflink>]; [<reflink idref="bib47" id="ref117">47</reflink>]). However, they are significantly lower than the correlations between two dependently coded occupations, which range from 0.97 to 0.98. Continuous measures seem to increase the reliability of occupational coding, but our results show that they do not solve the measurement error problem in estimating occupational mobility.</p> <hd id="AN0192656321-17">Robustness Checks</hd> <p>For the occupational categories, we use the IPUMS 1990 classifications. To check whether our results vary depending on which occupational classification systems are used, we repeat the same analyses using the original non-harmonized occupations and the IPUMS 2010 occupational classification, getting nearly identical results as we report here. We also repeat our analyses with the recoded 1990 occupational categories of [<reflink idref="bib2" id="ref118">2</reflink>] as well as the harmonized 2010 occupational categories of [<reflink idref="bib43" id="ref119">43</reflink>]. Our conclusions are not altered.</p> <p>In studies of occupational mobility, micro-classes ([<reflink idref="bib67" id="ref120">67</reflink>]) and EGP classes ([<reflink idref="bib16" id="ref121">16</reflink>]) are frequently used. Although we have tested both three-digit and one-digit occupations, some may be interested in testing whether these class schemes yield different results. As a further robustness check, we repeat the same analyses using these classes and find that the results do not change (see Supplemental Appendix Figure A.3). The mismatch rates with 82 micro-classes are around midway between those with three-digit and one-digit occupations. The mismatch rates with 10 EGP classes are slightly higher than those with one-digit occupations.</p> <p>During the study period, the Census Bureau made a major update to the occupation classification system and implemented it in the CPS in 2003. There are two more updates: one in 2011 and the other in 2020, but these updates are relatively minor compared to the 2003 update. We check whether the results change if the data before 2003 are excluded. As shown in Figure 4, the predicted probabilities of occupational mobility continue to rise after 2003, and the divergence between the mobility measures is again evident when 2003 is set to zero. We also repeat the same decomposition analyses as in Table 5 using the years 2004–2007 instead of 1992–1996. Our conclusions are not altered.</p> <p>Including respondents whose age, gender, and race/ethnicity do not match in two adjacent years does not change our conclusions either, as the proportion of those mismatches is stable at around 2 percent to 3 percent for age and around 1 percent for gender and race. The mobility estimates of <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mi>D</mi></msup></math> </ephtml> , <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><msup><mi>D</mi><mrow><mi mathvariant="normal">′</mi></mrow></msup></mrow></msup></math> </ephtml> , and <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>p</mi></mrow></msup></math> </ephtml> need the CPS-ASEC, but <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> can be estimated with the basic monthly CPS. Using other months to calculate <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><msup><mi>m</mi><mrow><mi>I</mi><mi>c</mi></mrow></msup></math> </ephtml> yields similarly high occupational mobilities as we report here.</p> <p>Some may suspect that the high coding mismatch rates in general and the rising mismatches over time in particular might be associated with proxy interviews. Indeed, proxy interviews are associated with slightly higher mismatches than self-responses (Table 3). However, the control of proxy interviews does not account for the rise in coding mismatches (Table 3). Instead, changes in proxy interviews between 1992–1996 and 2015–2019 led to a decline in coding mismatches rather than a rise (Table 5). Nevertheless, as a robustness check, we replicate Figure 3 after dropping all cases with proxy reports to see if the occupational mobility patterns without proxy interviews differ from those with proxy interviews, finding that the patterns with and without proxy interviews are almost identical (see Supplemental Appendix Figure A4).</p> <hd id="AN0192656321-18">Discussion and Conclusion</hd> <p>Exploiting the rotating panel design of the CPS, this study explored the coding mismatch between current and previous occupations and its impact on measures of occupational career mobility. We compute four measures of occupational mobility between two adjacent years. Ideally, these measures should yield very similar mobilities. Our results, however, show that the discrepancies across measures are astonishingly large. Around 45 percent of the current occupations answered in the year <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo><mn>1</mn></math> </ephtml> do not match the previous occupations answered in the year <emph>t</emph> by the same respondents. The conceptual difference between current and previous occupation does not seem to be associated with the large discrepancy. Recall errors may account for, at best, only a small portion of the discrepancies. Nor do respondent characteristics associated with response reliability account for the large discrepancies, even though respondents' demographic covariates are associated with the likelihood of mismatches. Our results show that low intercoder reliability is likely the main cause of the mismatch. We also find that the mismatch has increased over time, even after controlling for changes in demographic covariates and occupational structure. The overestimation of occupational mobility based on two independently coded occupations has worsened in recent years.</p> <p>Sociologists have argued that occupation is a more reliable measure of socioeconomic status than other measures such as income because it is less prone to measurement error and easy to recall ([<reflink idref="bib21" id="ref122">21</reflink>]; [<reflink idref="bib64" id="ref123">64</reflink>]). Recent studies, however, have cast doubt on this claim. Annual income is a better proxy for lifetime earnings than detailed occupations ([<reflink idref="bib40" id="ref124">40</reflink>]; [<reflink idref="bib56" id="ref125">56</reflink>]). Establishments and jobs account for more variance in wages than occupations ([<reflink idref="bib3" id="ref126">3</reflink>]). The mismatch between coders for the same occupation description is quite common ([<reflink idref="bib12" id="ref127">12</reflink>]; [<reflink idref="bib5" id="ref128">5</reflink>]; [<reflink idref="bib40" id="ref129">40</reflink>]). The substantial variation in the description of activities and duties within the same detailed occupation is widespread ([<reflink idref="bib46" id="ref130">46</reflink>]). There seems to be a growing awareness of the limitations of the occupation variable ([<reflink idref="bib54" id="ref131">54</reflink>]). Our study adds to the growing literature by presenting the case using the CPS.</p> <p>The classification of occupations by coders and the boundary-making of occupations are socially constructed. By asking an open-ended query on job titles, the State of Texas collected more than half a million job titles in its job assessment database ([<reflink idref="bib63" id="ref132">63</reflink>]). The activities and duties of an occupation consist of different types of responsibilities and functions shared across multiple occupations. An occupation is a collection of jobs that are heterogeneous in their tasks. Although occupation is a discrete variable, many tasks, activities, and duties of an occupation are not confined within one occupation ([<reflink idref="bib24" id="ref133">24</reflink>]). Thus, coding mismatches are an inherent reflection of an occupation variable.</p> <p>Faced with this reliability problem in occupational coding, government statistical agencies, including the US Census Bureau, have tried to develop rule-based occupational coding (e.g., <emph>The Alphabetic Indexes of Occupations</emph> ([<reflink idref="bib10" id="ref134">10</reflink>])) by which the presence of a particular word should be matched with a particular occupation. They have also provided computer-assisted structural coding tools (e.g., the ACS Industry and Occupation (I&O) Autocoder) ([<reflink idref="bib19" id="ref135">19</reflink>]; [<reflink idref="bib28" id="ref136">28</reflink>]; [<reflink idref="bib55" id="ref137">55</reflink>]). These tools aim to assist professional coders rather than replace them. For example, the ACS I&O Autocoder reduces the burden on professional coders by automating relatively obvious cases ([<reflink idref="bib61" id="ref138">61</reflink>]). The target intercoder reliability of the ACS I&O Autocoder is to reach the level of human coding because professional coders are more accurate than autocoding ([<reflink idref="bib61" id="ref139">61</reflink>]). Despite efforts to automate, coding of occupations has always involved the subjective judgments of coders.</p> <p>Some researchers have tried to develop a complicated procedure such as a multi-trait multi-method (MTMM) model in occupational coding (e.g., [<reflink idref="bib18" id="ref140">18</reflink>]). Such efforts are valuable and would improve the reliability of occupation, but they undermine one of the rationales for using occupation in stratification studies, which is simplicity and ease of measurement. Even without such a complicated model, collecting reliable occupation information is hard and expensive. The added benefit of making occupation coding more complicated is questionable. AI-assisted coding with pre-trained language models can be suggested as a solution ([<reflink idref="bib41" id="ref141">41</reflink>]). For example, by analyzing a German survey dataset, [<reflink idref="bib53" id="ref142">53</reflink>] reported that AI-assisted coding is superior to algorithm-based coding in terms of reliability. The reliability of AI-assisted coding can be improved with rigorous language training ([<reflink idref="bib41" id="ref143">41</reflink>]). A potential issue is the resources for language training. The Census Bureau needs to update occupational classification schemes because the occupational structure is constantly changing with the development of the New Economy. To date, human coders have determined the classification of new occupations. As the number of clerical coding decreases, the resources for language training for AI may diminish. It is too early to judge the quality of AI-assisted occupational coding and related issues. Future research on this topic is warranted.</p> <p>Although coding mismatches between coders per se are not measurement errors, they result in measurement errors in occupational mobility. Occupational mobility between two points in time requires two descriptions of the occupation and the involvement of (usually) two coders. The reliability problem of independently coded occupations encompasses both intercoder and response reliability. Efforts by government statistical agencies have focused on intercoder reliability and response reliability seems to have been ignored. To our knowledge, there has been no single study of response reliability. Our results suggest that response reliability may not be the main source of mismatches, but we cannot completely rule out the possibility that response reliability is a significant source of mismatches. [<reflink idref="bib46" id="ref144">46</reflink>] reported that the descriptions of the same occupation by different respondents vary, but we do not know how the same respondent's descriptions of the same occupation vary when answered in two distant time points. Future research on response reliability is needed.</p> <p>Our results caution researchers studying intergenerational mobility using the linked census data or surveys where occupations of parents and children are asked independently. It is tempting to assume that occupations answered independently by parents and children would be more accurate than those answered only by children. However, [<reflink idref="bib15" id="ref145">15</reflink>] found that when children describe both their parents' and their own occupations, the match rate between different coders tends to be higher for the parents' occupation than for the respondent's current occupation. This implies that using data in which parents and children answer their occupation independently does not necessarily offer more accurate estimates of intergenerational mobility than data in which children provide all the information. As full-count census data become available and IPUMS offers a convenient linking program ([<reflink idref="bib23" id="ref146">23</reflink>]), scholars have begun to study intergenerational occupational mobility using these datasets (e.g., [<reflink idref="bib58" id="ref147">58</reflink>]). Our results imply that the estimation of intergenerational mobility based on these data can be upwardly biased. Coding mismatches are not confined to detailed occupations, so broad occupational classifications do not solve the problem.</p> <p>Thus, we echo [<reflink idref="bib54" id="ref148">54</reflink>]) call for using earnings and wages rather than occupation in the study of intergenerational mobility. We support the same argument for intragenerational mobility. Contrary to the widespread assumption in sociology, occupation information is not easy to gather; coding occupation is time-consuming, expensive, and error-prone. The validity and reliability of occupation are less understood than earnings. As high-quality earnings and wage information become increasingly available, sociologists studying mobility can benefit from using earnings and wages rather than occupation variables. As for inequality-generating mechanisms, we echo [<reflink idref="bib3" id="ref149">3</reflink>] that establishment- or job-centered research rather than occupation-centered agenda is needed.[<reflink idref="bib15" id="ref150">15</reflink>]</p> <p>As a relatively easy fix to the CPS, we recommend the introduction of the dependent coding in the fifth month of the CPS. To reduce mismatches, coders must have access to the descriptions of occupations in the previous round of the survey. In the CPS-ASEC, dependent coding was introduced in 1970 in comparing the current job with the longest job held last year. A similar dependent coding technique was implemented in 1994 in the monthly CPS, through which month-to-month occupation changes are measured. However, in the fifth month, which is the month the respondents are re-interviewed after the eight-month break, occupations are independently coded without referring to the occupation held by the respondent in the fourth month. We recommend that the Census Bureau consider implementing dependent coding techniques in the fifth month. The need for dependent coding in the fifth month is strong, given the increase in coding mismatches over the past three decades.</p> <p>A main limitation of the current study is that our results do not offer a clear explanation for why coding mismatches have increased other than that coding mismatches between coders might have increased. We suspect that changes in coding procedures and/or turnovers in professional coders might be associated with the rise. Further research on this topic is needed.</p> <hd id="AN0192656321-19">Supplemental Material</hd> <p>Graph: Supplemental material, sj-pdf-1-smr-10.1177_00491241241303517 for The Rise in Occupational Coding Mismatches and Occupational Mobility, 1991–2020 by Andrew Taeho Kim and Chang Hwan Kim in Sociological Methods & Research</p> <hd id="AN0192656321-20">Acknowledgements</hd> <p>We thank the Editor and the three anonymous reviewers for their invaluable and constructive comments. We also benefited greatly from the feedback provided by Xi Song, Siwei Cheng, Donna Ginther, and the CPS team at the U.S. Census Bureau. Misty Heggeness connected us to the CPS team at the U.S. Census Bureau. An earlier version of this paper was presented at the 2022 Southern Demographic Association annual meeting and the 2023 Population Association of America annual meeting.</p> <ref id="AN0192656321-21"> <title> References </title> <blist> <bibl id="bib1" idref="ref77" type="bt">1</bibl> <bibtext> Abrevaya J., Hausman J. A. 1999. " Semiparametric Estimation with Mismeasured Dependent Variables: An Application to Duration Models for Unemployment Spells." Annales d'Économie et de Statistique 55/56: 243–75.</bibtext> </blist> <blist> <bibl id="bib2" idref="ref83" type="bt">2</bibl> <bibtext> Autor D. H., Dorn D. 2013. " The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market." American Economic Review. 103: 1553–97.</bibtext> </blist> <blist> <bibl id="bib3" idref="ref84" type="bt">3</bibl> <bibtext> Avent-Holt D., Henriksen L. F., Hägglund A. E., Jung J., Kodama N., Melzer S. M., Mun E., Rainey A., Tomaskovic-Devey D. 2020. " Occupations, Workplaces or Jobs? An Exploration of Stratification Contexts Using Administrative Data." Research in Social Stratification and Mobility. 70: 100456.</bibtext> </blist> <blist> <bibl id="bib4" idref="ref38" type="bt">4</bibl> <bibtext> Babbie E. R.2020. The Practice of Social Research. 15th ed. Boston, MA: Cengage.</bibtext> </blist> <blist> <bibl id="bib5" idref="ref12" type="bt">5</bibl> <bibtext> Belloni M., Brugiavini A., Meschi E., Tijdens K. 2016. " Measuring and Detecting Errors in Occupational Coding: An Analysis of SHARE Data." Journal of Official Statistics. 32: 917–45.</bibtext> </blist> <blist> <bibl id="bib6" idref="ref59" type="bt">6</bibl> <bibtext> Bernhardt R., Munro D., Wolcott E. 2021. " How Does the Dramatic Rise of CPS Non-Response Impact Labor Market Indicators?" GLO Discussion Paper Series781.</bibtext> </blist> <blist> <bibl id="bib7" idref="ref39" type="bt">7</bibl> <bibtext> Biemer P. P., Lyberg L. E. 2003. Introduction to Survey Quality. Hoboken, NJ: Wiley & Sons, Inc.</bibtext> </blist> <blist> <bibl id="bib8" idref="ref2" type="bt">8</bibl> <bibtext> Blau P. M., Duncan O. D. 1967. The American Occupational Structure. New York, NY: Wiley.</bibtext> </blist> <blist> <bibl id="bib9" idref="ref46" type="bt">9</bibl> <bibtext> Campanelli P. 1997. " Testing Survey Questions: New Directions in Cognitive Interviewing." Bulletin of Sociological Methodology. 55: 5–17.</bibtext> </blist> <blist> <bibtext> Census. 2019. " The Alphabetical Indexes of Industries and Occupations. " United States Census Bureau.</bibtext> </blist> <blist> <bibtext> Cheng S., Park B. 2020. " Flows and Boundaries: A Network Approach to Studying Occupational Mobility in the Labor Market." American Journal of Sociology. 126: 577–631.</bibtext> </blist> <blist> <bibtext> Conrad F. G., Couper M. P., Sakshaug J. W.2016. " Classifying Open-Ended Reports: Factors Affecting the Reliability of Occupation Codes." Journal of Official Statistics. 32: 75–92.</bibtext> </blist> <blist> <bibtext> Duncan O. D.1961. "A Socioeconomic Index for All Occupations." Pp.109–138 in Occupations and Social Status, edited by Reis A. J. New York, NY: Free Press of Glencoe.</bibtext> </blist> <blist> <bibtext> Elias P.1997. "Occupational Classification (ISCO-88): Concepts, Methods, Reliability, Validity and Cross-National Comparability. OECD Labour Market and Social Policy Occasional Papers 20.</bibtext> </blist> <blist> <bibtext> Ellison R. 2014. " CASCOT and the Coding of Occupations in European Surveys. " Technical report, Amsterdam, the Netherlands.</bibtext> </blist> <blist> <bibtext> Erikson R., Goldthorpe J. H., Portocarero L. 1979. " Intergenerational Class Mobility in Three Western European Societies: England, France and Sweden." British Journal of Sociology. 30: 415–41.</bibtext> </blist> <blist> <bibtext> Flood S., King M., Rodgers R., Ruggles S., Warren J.R. 2020. "Integrated Public Use Microdata Series, Current Population Survey: Version 8.0 [Dataset]." Minneapolis, MN: University of Minnesota.</bibtext> </blist> <blist> <bibtext> Ganzeboom H. B. G. 2010. "With Special Reference to the International Standard Classification of Occupations ISCO-88 with an Extension on ISCO-08." Working Paper.</bibtext> </blist> <blist> <bibtext> Gweon H., Schonlau M., Kaczmirek L., Blohm M., Steiner S. 2017. " Three Methods for Occupation Coding Based on Statistical Learning." Journal of Official Statistics. 33: 101–22.</bibtext> </blist> <blist> <bibtext> Halpern-Manners A., Warren J. R. 2012. " Panel Conditioning in Longitudinal Studies: Evidence from Labor Force Items in the Current Population Survey." Demography. 49: 1499–519.</bibtext> </blist> <blist> <bibtext> Hauser R. M., Warren J. R. 1997. " Socioeconomic Indexes for Occupations: A Review, Update, and Critique." Sociological Methodology. 27: 177–298.</bibtext> </blist> <blist> <bibtext> Hausman J. A., Abrevaya J., Scott-Morton F. M. 1998. " Misclassification of the Dependent Variable in a Discrete-Response Setting." Journal of Econometrics. 87: 239–69.</bibtext> </blist> <blist> <bibtext> Helgertz S. R., Warren J. R., Fitch C. A., Goeken R., Hacker J. D., Nelson M. A., Price J. P., Roberts E., Sobek M. 2020. " IPUMS Multigenerational Longitudinal Panel: Version 1.0 [Dataset]. " Minneapolis, MN: University of Minnesota. https://doi.org/10.18128/D016.V1.0.</bibtext> </blist> <blist> <bibtext> Hénaut L., Lena J. C., Accominotti F. 2023. " Polyoccupationalism: Expertise Stretch and Status Stretch in the Postindustrial Era." American Sociological Review. 88: 872–900.</bibtext> </blist> <blist> <bibtext> Hokayem C., Bollinger C., Ziliak J. P.2015. " The Role of CPS Nonresponse in the Measurement of Poverty." Journal of the American Statistical Association. 110: 935–45.</bibtext> </blist> <blist> <bibtext> Houseworth C., Fisher J. 2020. " Measurement Error in Occupation and the Impact on Intergenerational Mobility." Research in Social Stratification and Mobility. 70: 100469.</bibtext> </blist> <blist> <bibtext> Hout M., Hastings O. P.2016. " Reliability of the Core Items in the General Social Survey: Estimates from the Three-Wave Panels, 2006–2014." Sociological Science. 3: 971–1002.</bibtext> </blist> <blist> <bibtext> Ikudo A., Lane J. I., Staudt J., Weinberg B. A.2019. " Occupational Classifications: A Machine Learning Approach." Journal of Economic and Social Measurement. 44: 57–87.</bibtext> </blist> <blist> <bibtext> International Labour Office. 2012. International Standard Classification of Occupations 2008 (ISCO-08): Structure, Group Definitions and Correspondence Tables. Geneva: International Labour Office.</bibtext> </blist> <blist> <bibtext> Jarvis B. F., Song X. 2017. " Rising Intragenerational Occupational Mobility in the United States, 1969 to 2011." American Sociological Review. 82: 568–99.</bibtext> </blist> <blist> <bibtext> Jerry H.2001. " Mismeasured Variables in Econometric Analysis: Problems from the Right and Problems from the Left." Journal of Economic Perspectives. 15: 57–67.</bibtext> </blist> <blist> <bibtext> Kalleberg A. L., Mouw T. 2018. " Occupations, Organizations, and Intragenerational Career Mobility." Annual Review of Sociology. 44: 283–303.</bibtext> </blist> <blist> <bibtext> Kambourov G., Manovskii I. 2008. " Rising Occupational and Industry Mobility in the United States: 1968–97." International Economic Review. 49: 41–79.</bibtext> </blist> <blist> <bibtext> Kambourov G., Manovskii I. 2013. " A Cautionary Note on Using (March) Current Population Survey and Panel Study of Income Dynamics Data to Study Worker Mobility." Macroeconomic Dynamics. 17: 172–94.</bibtext> </blist> <blist> <bibtext> Kan M. Y., Pudney S.2008. " Measurement Error in Stylized and Diary Data on Time Use." Sociological Methodology. 38: 101–32.</bibtext> </blist> <blist> <bibtext> Kim C. H.2013a. " Detailed Wage Decompositions: Revisiting the Identification Problem." Sociological Methodology. 43: 346–63.</bibtext> </blist> <blist> <bibtext> Kim Y.-M.2013b. " Diverging Top and Converging Bottom: Labour Flexibilization and Changes in Career Mobility in the USA." Work, Employment and Society. 27: 860–79.</bibtext> </blist> <blist> <bibtext> Kim C. H., Kim J., Ban M. 2020. " Do You Know What You Do for a Living? Occupational Coding Mismatches between Coders in the Korean General Social Survey." Research in Social Stratification and Mobility. 70: 100467.</bibtext> </blist> <blist> <bibtext> Kim C. H., Tamborini C. R.2014. " Response Error in Earnings: An Analysis of the Survey of Income and Program Participation Matched with Administrative Data." Sociological Methods & Research. 43: 39–72.</bibtext> </blist> <blist> <bibtext> Kim C. H., Tamborini C. R., Sakamoto A. 2018. " The Sources of Life Chances: Does Education, Class Category, Occupation, or Short-Term Earnings Predict 20-Year Long-Term Earnings?" Sociological Science. 5: 206–33.</bibtext> </blist> <blist> <bibtext> Laughlin L., Song X., Wisniewski M., Xu J. 2024. " From Job Descriptions to Occupations: Using LLMs to Code and Rank Occupations. " in Presentation at the Annual Meeting of the Population Association.</bibtext> </blist> <blist> <bibtext> Leicht K. T.2020. " Occupations and Inequalities in the 21st Century: What's in Your Wallet?" Research in Social Stratification and Mobility. 70: 100550.</bibtext> </blist> <blist> <bibtext> Lin K.-H., Hung K. 2022. " The Network Structure of Occupations: Fragmentation, Differentiation, and Contagion." American Journal of Sociology. 127: 1551–601.</bibtext> </blist> <blist> <bibtext> Mannetje A., Kromhout H.2003. " The Use of Occupation and Industry Classifications in General Population Studies." International Journal of Epidemiology. 32: 419–28.</bibtext> </blist> <blist> <bibtext> Manzoni A., Vermunt J. K., Luijkx R., Muffels R. 2010. " Memory Bias in Retrospectively Collected Employment Careers: A Model-Based Approach to Correct for Measurement Error." Sociological Methodology. 40: 39–73.</bibtext> </blist> <blist> <bibtext> Martin-Caughey A.2021. " What's in an Occupation? Investigating Within-Occupation Variation and Gender Segregation Using Job Titles and Task Descriptions." American Sociological Review. 86: 960–99.</bibtext> </blist> <blist> <bibtext> Massing N., Wasmer M., Wolf C., Zuell C. 2019. " How Standardized is Occupational Coding? A Comparison of Results from Different Coding Agencies in Germany." Journal of Official Statistics. 35: 167–87.</bibtext> </blist> <blist> <bibtext> Mathiowetz N. A. 1992. " Errors in Reports of Occupation." Public Opinion Quarterly. 56: 352–5.</bibtext> </blist> <blist> <bibtext> Mellow W., Sider H. 1983. " Accuracy of Response in Labor Market Surveys: Evidence and Implications." Journal of Labor Economics. 1: 331–44.</bibtext> </blist> <blist> <bibtext> Nakao K., Treas J. 1994. " Updating Occupational Prestige and Socioeconomic Scores: How the New Measures Measure Up." Sociological Methodology. 24: 1–72.</bibtext> </blist> <blist> <bibtext> Phelan J. C., Link B. G., Tehranifar P. 2010. " Social Conditions as Fundamental Causes of Health Inequalities: Theory, Evidence, and Policy Implications." Journal of Health and Social Behavior. 51: S28–S40.</bibtext> </blist> <blist> <bibtext> Poterba J. M., Summers L. H.1995. " Unemployment Benefits and Labor Market Transitions: A Multinomial Logit Model with Errors in Classification." Review of Economics & Statistics. 77: 207–16.</bibtext> </blist> <blist> <bibtext> Safikhani P., Avetisyan H., Föste-Eggers D., Broneske D. 2023. " Automated Occupation Coding with Hierarchical Features: A Data-Centric Approach to Classification with Pre-Trained Language Models." Discover Artificial Intelligence. 3: 6.</bibtext> </blist> <blist> <bibtext> Sakamoto A., Wang S. X.2020. " The Declining Significance of Occupation in Research on Intergenerational Mobility." Research in Social Stratification and Mobility. 70: 100521.</bibtext> </blist> <blist> <bibtext> Schierholz M., Schonlau M. 2021. " Machine Learning for Occupation Coding—A Comparison Study." Journal of Survey Statistics and Methodology. 9: 1013–34.</bibtext> </blist> <blist> <bibtext> Shahbazian R., Bihagen E. 2022. " Does Your Class Give More than a Hint of Your Lifetime Earnings?: Assessing Indicators for Lifetime Earnings Over the Life Course for Sweden." European Sociological Review. 38: 527–42.</bibtext> </blist> <blist> <bibtext> Siegel P. M. (1971. " Prestige in the American Occupational Structure. " Ph.D. dissertation. Chicago, IL: University of Chicago.</bibtext> </blist> <blist> <bibtext> Song X., Massey C. G., Rolf K. A., Ferrie J. P., Rothbaum J. L., Xie Y. 2020. " Long-Term Decline in Intergenerational Mobility in the United States since the 1850s." Proceedings of the National Academy of Sciences of the United States of America. 117: 251–8.</bibtext> </blist> <blist> <bibtext> Song X., Xie Y. 2023. " Occupational Percentile Rank: A New Method for Constructing a Socioeconomic Index of Occupational Status." Sociological Methods & Research. doi: https://doi.org/10.1177/00491241231207914.</bibtext> </blist> <blist> <bibtext> Tamborini C. R., Kim C. H. 2013. " Are Proxy Interviews Associated with Biased Earnings Reports? Marital Status and Gender Effects of Proxy." Social Science Research. 42: 499–512.</bibtext> </blist> <blist> <bibtext> Thompson M., Kornbau M. E, Vesely J. 2012. " Creating an Automated Industry and Occupation Coding Process for the American Community Survey. " Census Working Papers.</bibtext> </blist> <blist> <bibtext> Tilly C. 1999. Durable Inequality. Berkeley, CA: University of California Press.</bibtext> </blist> <blist> <bibtext> Tippins N. T., Hilton M. L. (eds.). 2010. A Database for a Changing Economy: Review of the Occupational Information Network (O*NET). Washington, D.C. National Academies Press.</bibtext> </blist> <blist> <bibtext> Torche F.2015. " Analyses of Intergenerational Mobility: An Interdisciplinary Review." ANNALS of the American Academy of Political and Social Science. 657: 37–62.</bibtext> </blist> <blist> <bibtext> Wardle J., Waller J., Jarvis M. J.2002. " Sex Differences in the Association of Socioeconomic Status with Obesity." American Journal of Public Health. 92: 1299–304.</bibtext> </blist> <blist> <bibtext> Warren J. R., Halpern-Manners A. 2012. " Panel Conditioning in Longitudinal Social Science Surveys." Sociological Methods & Research. 41: 491–534.</bibtext> </blist> <blist> <bibtext> Weeden K. A., Grusky D. B.2005. " The Case for a New Class Map." American Journal of Sociology. 111: 141–212.</bibtext> </blist> <blist> <bibtext> Wright E. O. (ed.). 2005. Approaches to Class Analysis. Cambridge, MA: Cambridge University Press.</bibtext> </blist> </ref> <ref id="AN0192656321-22"> <title> Footnotes </title> <blist> <bibtext> The IPUMS CPS variable names and the Stata codes to replicate the results are available at an OSF website: https://osf.io/2u8rx/.</bibtext> </blist> <blist> <bibtext> The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.</bibtext> </blist> <blist> <bibtext> The author(s) received no financial support for the research, authorship, and/or publication of this article.</bibtext> </blist> <blist> <bibtext> Andrew Taeho Kim https://orcid.org/0000-0002-1402-2927 ChangHwan Kim https://orcid.org/0000-0001-7149-1386</bibtext> </blist> <blist> <bibtext> The raw data of this study are available in the IPUMS CPS at https://cps.ipums.org/cps/.</bibtext> </blist> <blist> <bibtext> Supplemental material for this article is available online.</bibtext> </blist> <blist> <bibtext> Among the target population of the current study, the non-response rates to questions about current and previous occupations remain below 1 percent for most years between 1991 and 2020.</bibtext> </blist> <blist> <bibtext> https://www..census.gov/programs-surveys/cps/methodology/Occupation%20Codes.pdf. Accessed March 12, 2024.</bibtext> </blist> <blist> <bibtext> We thank Reviewer 3 for introducing <emph>The Alphabetic Indexes of Industry and Occupation</emph>.</bibtext> </blist> <blist> <bibtext> [34]: 182–183) argued that the annual occupational mobility of the March CPS, which compares the current occupation and the occupation in the previous year, is downwardly biased as it is close to the two and three-month mobility found in the monthly CPS.</bibtext> </blist> <blist> <bibtext> The list of other supplemental CPS surveys can be found at https://<ulink href="http://www.census.gov/programs-surveys/cps/about/supplemental-surveys.html">www.census.gov/programs-surveys/cps/about/supplemental-surveys.html</ulink>.</bibtext> </blist> <blist> <bibtext> Some may worry that panel conditioning can bias the results of longitudinal surveys ([20]; [66]). However, because our study compares various occupational mobility measures between <ephtml> <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>t</mi><mo>−</mo>1</math> </ephtml> and <emph>t</emph> for the same respondents rather than trying to estimate true occupational mobility, our conclusions are unlikely to be affected by panel conditioning problems.</bibtext> </blist> <blist> <bibtext> Some may suspect that the increase in multiple jobs might account for the growth in coding mismatches. The monthly CPS has information to identify multiple jobs, but the oversampled respondents in the CPS-ASEC do not have such information. We limit our sample to those who can be linked to the March CPS and conduct additional analyses with the control of the indicators of multiple jobs (not shown here). Our conclusions are unchanged.</bibtext> </blist> <blist> <bibtext> Using the SEI ([21]) or two versions of the occupational prestige score ([50]; [57]) yields nearly identical results with what we report here.</bibtext> </blist> <blist> <bibtext> The presented results and discussion do not deny the usefulness of occupation in some domains. The occupational gradient in health outcomes is a prime example (e.g., [51]; [65]). However, even in these cases, scholars need to be aware of the substantial limitations of occupation variables and use them with caution.</bibtext> </blist> </ref> <aug> <p>By Andrew Taeho Kim and ChangHwan Kim</p> <p>Reported by Author; Author</p> <p></p> <p>Andrew Taeho Kim is an assistant professor of sociology at the University of Tennessee, Knoxville. His research interests include labor markets, stratification and inequality, race and gender in the labor market, Asian Americans, and quantitative methodology.</p> <p>ChangHwan Kim is a professor of sociology at the University of Kansas. His research focuses on labor markets, education, Asian American studies, and Korean studies. Methodologically, he specializes in panel decomposition and measurement error analysis.</p> </aug> <nolink nlid="nl1" bibid="bib62" firstref="ref1"></nolink> <nolink nlid="nl2" bibid="bib64" firstref="ref3"></nolink> <nolink nlid="nl3" bibid="bib16" firstref="ref4"></nolink> <nolink nlid="nl4" bibid="bib68" firstref="ref5"></nolink> <nolink nlid="nl5" bibid="bib67" firstref="ref6"></nolink> <nolink nlid="nl6" bibid="bib11" firstref="ref7"></nolink> <nolink nlid="nl7" bibid="bib30" firstref="ref8"></nolink> <nolink nlid="nl8" bibid="bib32" firstref="ref9"></nolink> <nolink nlid="nl9" bibid="bib21" firstref="ref10"></nolink> <nolink nlid="nl10" bibid="bib12" firstref="ref13"></nolink> <nolink nlid="nl11" bibid="bib26" firstref="ref14"></nolink> <nolink nlid="nl12" bibid="bib34" firstref="ref15"></nolink> <nolink nlid="nl13" bibid="bib38" firstref="ref16"></nolink> <nolink nlid="nl14" bibid="bib42" firstref="ref17"></nolink> <nolink nlid="nl15" bibid="bib46" firstref="ref18"></nolink> <nolink nlid="nl16" bibid="bib54" firstref="ref19"></nolink> <nolink nlid="nl17" bibid="bib48" firstref="ref23"></nolink> <nolink nlid="nl18" bibid="bib49" firstref="ref24"></nolink> <nolink nlid="nl19" bibid="bib25" firstref="ref31"></nolink> <nolink nlid="nl20" bibid="bib33" firstref="ref33"></nolink> <nolink nlid="nl21" bibid="bib37" firstref="ref34"></nolink> <nolink nlid="nl22" bibid="bib29" firstref="ref35"></nolink> <nolink nlid="nl23" bibid="bib24" firstref="ref36"></nolink> <nolink nlid="nl24" bibid="bib63" firstref="ref43"></nolink> <nolink nlid="nl25" bibid="bib14" firstref="ref48"></nolink> <nolink nlid="nl26" bibid="bib44" firstref="ref50"></nolink> <nolink nlid="nl27" bibid="bib39" firstref="ref60"></nolink> <nolink nlid="nl28" bibid="bib22" firstref="ref62"></nolink> <nolink nlid="nl29" bibid="bib52" firstref="ref63"></nolink> <nolink nlid="nl30" bibid="bib31" firstref="ref64"></nolink> <nolink nlid="nl31" bibid="bib10" firstref="ref66"></nolink> <nolink nlid="nl32" bibid="bib27" firstref="ref74"></nolink> <nolink nlid="nl33" bibid="bib58" firstref="ref79"></nolink> <nolink nlid="nl34" bibid="bib59" firstref="ref80"></nolink> <nolink nlid="nl35" bibid="bib35" firstref="ref88"></nolink> <nolink nlid="nl36" bibid="bib45" firstref="ref90"></nolink> <nolink nlid="nl37" bibid="bib60" firstref="ref93"></nolink> <nolink nlid="nl38" bibid="bib17" firstref="ref94"></nolink> <nolink nlid="nl39" bibid="bib13" firstref="ref103"></nolink> <nolink nlid="nl40" bibid="bib36" firstref="ref104"></nolink> <nolink nlid="nl41" bibid="bib47" firstref="ref111"></nolink> <nolink nlid="nl42" bibid="bib43" firstref="ref119"></nolink> <nolink nlid="nl43" bibid="bib40" firstref="ref124"></nolink> <nolink nlid="nl44" bibid="bib56" firstref="ref125"></nolink> <nolink nlid="nl45" bibid="bib19" firstref="ref135"></nolink> <nolink nlid="nl46" bibid="bib28" firstref="ref136"></nolink> <nolink nlid="nl47" bibid="bib55" firstref="ref137"></nolink> <nolink nlid="nl48" bibid="bib61" firstref="ref138"></nolink> <nolink nlid="nl49" bibid="bib18" firstref="ref140"></nolink> <nolink nlid="nl50" bibid="bib41" firstref="ref141"></nolink> <nolink nlid="nl51" bibid="bib53" firstref="ref142"></nolink> <nolink nlid="nl52" bibid="bib15" firstref="ref145"></nolink> <nolink nlid="nl53" bibid="bib23" firstref="ref146"></nolink>
Header DbId: eric
DbLabel: ERIC
An: EJ1502110
AccessLevel: 3
PubType: Academic Journal
PubTypeId: academicJournal
PreciseRelevancyScore: 0
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: The Rise in Occupational Coding Mismatches and Occupational Mobility, 1991-2020
– Name: Language
  Label: Language
  Group: Lang
  Data: English
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AR" term="%22Andrew+Taeho+Kim%22">Andrew Taeho Kim</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0002-1402-2927">0000-0002-1402-2927</externalLink>)<br /><searchLink fieldCode="AR" term="%22ChangHwan+Kim%22">ChangHwan Kim</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0001-7149-1386">0000-0001-7149-1386</externalLink>)
– Name: TitleSource
  Label: Source
  Group: Src
  Data: <searchLink fieldCode="SO" term="%22Sociological+Methods+%26+Research%22"><i>Sociological Methods & Research</i></searchLink>. 2026 55(2):659-699.
– Name: Avail
  Label: Availability
  Group: Avail
  Data: SAGE Publications. 2455 Teller Road, Thousand Oaks, CA 91320. Tel: 800-818-7243; Tel: 805-499-9774; Fax: 800-583-2665; e-mail: journals@sagepub.com; Web site: https://sagepub.com
– Name: PeerReviewed
  Label: Peer Reviewed
  Group: SrcInfo
  Data: Y
– Name: Pages
  Label: Page Count
  Group: Src
  Data: 41
– Name: DatePubCY
  Label: Publication Date
  Group: Date
  Data: 2026
– Name: TypeDocument
  Label: Document Type
  Group: TypDoc
  Data: Journal Articles<br />Reports - Research
– Name: Subject
  Label: Descriptors
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22Occupational+Mobility%22">Occupational Mobility</searchLink><br /><searchLink fieldCode="DE" term="%22Coding%22">Coding</searchLink><br /><searchLink fieldCode="DE" term="%22Occupations%22">Occupations</searchLink><br /><searchLink fieldCode="DE" term="%22National+Surveys%22">National Surveys</searchLink><br /><searchLink fieldCode="DE" term="%22Reliability%22">Reliability</searchLink>
– Name: SubjectThesaurus
  Label: Assessment and Survey Identifiers
  Group: Su
  Data: <searchLink fieldCode="SU" term="%22Current+Population+Survey%22">Current Population Survey</searchLink>
– Name: DOI
  Label: DOI
  Group: ID
  Data: 10.1177/00491241241303517
– Name: ISSN
  Label: ISSN
  Group: ISSN
  Data: 0049-1241<br />1552-8294
– Name: Abstract
  Label: Abstract
  Group: Ab
  Data: Occupation is a construct prone to classification mismatches by coders and description inconsistency by respondents. We explore whether mismatches in occupational coding have recently increased, what factors are associated with the rise in mismatches, and how the rise affects estimates of intragenerational occupational mobility. Utilizing the 1991-2020 Annual Social and Economic Supplement of the Current Population Survey, which collects information on respondents' current occupation and the previous year's main occupation, we identify coding mismatches and compare the probabilities of occupational mobility based on four combinations of two variables. Our results show that not only do the estimates of occupational mobility between two adjacent years vary substantially across measures, but also that the magnitudes of intragenerational occupational mobility across measures become increasingly decoupled over time. We demonstrate that the likely cause of this divergence is the rise in coding mismatches between coders. We discuss the implications of our findings.
– Name: AbstractInfo
  Label: Abstractor
  Group: Ab
  Data: As Provided
– Name: DateEntry
  Label: Entry Date
  Group: Date
  Data: 2026
– Name: AN
  Label: Accession Number
  Group: ID
  Data: EJ1502110
PLink https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=eric&AN=EJ1502110
RecordInfo BibRecord:
  BibEntity:
    Identifiers:
      – Type: doi
        Value: 10.1177/00491241241303517
    Languages:
      – Text: English
    PhysicalDescription:
      Pagination:
        PageCount: 41
        StartPage: 659
    Subjects:
      – SubjectFull: Occupational Mobility
        Type: general
      – SubjectFull: Coding
        Type: general
      – SubjectFull: Occupations
        Type: general
      – SubjectFull: National Surveys
        Type: general
      – SubjectFull: Reliability
        Type: general
      – SubjectFull: Current Population Survey
        Type: general
    Titles:
      – TitleFull: The Rise in Occupational Coding Mismatches and Occupational Mobility, 1991-2020
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: Andrew Taeho Kim
      – PersonEntity:
          Name:
            NameFull: ChangHwan Kim
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 05
              Type: published
              Y: 2026
          Identifiers:
            – Type: issn-print
              Value: 0049-1241
            – Type: issn-electronic
              Value: 1552-8294
          Numbering:
            – Type: volume
              Value: 55
            – Type: issue
              Value: 2
          Titles:
            – TitleFull: Sociological Methods & Research
              Type: main
ResultId 1