Comparability of categorical variables in longitudinal survey research Paul Lambert*, Alison Bowes*, Vernon Gayle*, Guy Warner*, and Tom Doherty** (*University of Stirling; ** University of Glasgow) Presented to the Society for Longitudinal & Life Course Studies Clare College, Cambridge 22-24 September 2010
29
Embed
Comparability of categorical variables in longitudinal survey research Paul Lambert*, Alison Bowes*, Vernon Gayle*, Guy Warner*, and Tom Doherty** (*University.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Comparability of categorical variables in longitudinal survey research
Paul Lambert*, Alison Bowes*, Vernon Gayle*, Guy Warner*, and Tom Doherty**
(*University of Stirling; ** University of Glasgow)
Presented to the Society for Longitudinal & Life Course StudiesClare College, Cambridge22-24 September 2010
Comparability of categorical variables in longitudinal survey research
1) Comparability strategies and selected problems
2) Some analytical prescriptions
3) The relevance of the ‘GESDE’ services
..Underlying motivation to study patterns/changes over time in distributions measured by categorical instruments..
“..Comparisons Are the Essence..” (Treiman, 2009: 382)
2
1) Comparability strategies
Harmonisation/standardisation is used to aid in the comparison of data across contextsMeasurement v’s meaning/functional equivalence
[i.e. same absolute qualities of position, v’s same relative meaning of position within the distribution]
Ex post v’s pre-harmonisation [e.g. Harkness et al. 2003; Hoffmeyer-Zlotnik & Harkness 2005]
Comparability problem: when comparison (over time) is muddled by a lack of equivalence in absolute and/or relative position
• Validity studies may be used to test correlation with expectations (e.g. Rose and Harrison 2010)
This is commonly visible as coding to the ‘lowest common denominator’
8
9
A measurement equivalence example – comparing ethnicity categories in context of demographic change
Are the compatible categories equivalent? Who does the work (data distributors and/or analysts)…? …& who records it (e.g. Mohler et al., 2008)?
10
Here, measurement equivalence is compromised by administrative errors, & meaning equivalence is doubtful due to industrial restructuring (orig. occ. codes not available)
Unskilled
Skilled manual
Petty-bourg.
Non-manual
Salariat
Source: Females from LFS/GHS, using data from Li and Heath (2008)
percent of year category
Goldthorpe class scheme harmonised over time
[Changing correlates] Occupational measures change over time in total distribution and this interacts with occupational gender segregation
11
0.1
.2.3
.4F
itted
va
lues
1990 1995 2000 2005 2010year
RGSC % advantaged EGP % advantaged
RGSC gender difference EGP gender difference
Source: BHPS waves 1-18, x-sectional respondent weighting, young adults 16-40 only
Gender trends 16-40yrs, social advantage in 2-category EGP and RGSC
…Meaning equivalence is preferable in longitudinal survey research…!!
Collect and preserve the data in its original detailPreserves information; easier for respondent; better documentation
Allow analysts to harmonise/standardise using flexible/tailored equivalence strategies
Same relative meaning may=diff. absolute properties (e.g. Educ-by-yob)?
Ideally, trying and comparing more than one strategy
Pressures for measurement equivalence from statistics agencies, analysts, consumers “..goal of standardisation is to enhance comparability; inappropriate
standardisation may do just the opposite” (Harkness, 2008: 61)
“..when nominal equivalence is not enough..” (Schneider, 2010) 12
2) Some analytical prescriptions
Measurement equivalence • ..is what we already do• Appealing because we have existing
standards/literature on most measures
Improved practice: • Use the published standards when they exist(!)
cf. temptation to recode/collapse recommended measures
• Documentation/replication of derivations• Comparison between a few different options• Recognise parsimony/multivariate interactions
13
14
.25
.3.3
5.4
.45
1990 2000 2010 2020 2030
Observed
Educ4 Educ4++Educ2a Educ2a++Educ2b Educ2b++
ISCED ISCED++
Source: Simulations on the BHPS balanced panel subsample (weighted by xlwght). Data from 1991-2007; simulations from 2008-2025. ++ lines show impact of educational upgradings.
Father-Child CAMSIS r, all adultsIntergenerational correlation simulations
(Own analysis, for the NeISS project, www.neiss.org.uk)
15
Meaning equivalence
For categorical data, equivalence for comparisons is often best approached in terms of meaning equivalence
(because of non-linear relations between categories and shifting underlying distributions)
(even if measurement equivalence seems possible)
Arithmetic standardisation offers a convenient form of meaning equivalence by indicating relative position with the structure defined by the current context
For categorical data, this can be achieved/approximated by scaling categories in one or more dimension of difference
16
None/low
Interm
FE/Voc
Univ
None/low
IntermFE/Voc
Univ 12
3
4567
89
1
2
3
4
567
89
3040
5060
70
FCS E L E L E L 60 70 80 90 00
Educ-4 scale scores Region averagesEGP class averages Occ major groups/orders
Individual FCS values (rescaled mean50, sd6)
Source: 'Slow Degrees' dataset 1963-2005 (Lambert et al., 2007), all adults 20+, N~=164k, categories are scaled by father's occ advantage (mean 50, sd15), E=1963-1989, L=1990-2005.
Effect proportional scaling using father's occupational advantage
Log likelihood = -47960.017 Prob > chi2 = 0.2616 Wald chi2(6) = 7.69Stereotype logistic regression Number of obs = 21999
Is scaling useful?
..sometimes.. Revealing exploratory exercise Parsimonious functional form
If complex categorical measure is a control variable If interaction effects are considered If story told by (transformation of) a linear functional
form is persuasive (e.g. exponential increase) Good for Multiple Response data
(e.g. take arithmetic function of related cases)
18
3) The GESDE services
DAMES (www.dames.org.uk) node seeking to support data management in social research activities
‘Data management’ = tasks involved in organising/ manipulating/enhancing data in preparation for research analysis Making tasks easier and more consistent
Pre-analysis tasks conducted by researchers and data distributors Re-coding; harmonising; standardising; cleaning; linking Recording and distributing metadata for replication
19
20
Contributions: Preserving/facilitating data management
Count
323 0 0 0 0 323
982 0 0 0 0 982
0 425 0 0 0 425
0 1597 0 0 0 1597
0 0 340 0 0 340
0 0 3434 0 0 3434
0 0 161 0 0 161
0 0 0 1811 0 1811
0 0 0 0 2518 2518
0 0 0 331 0 331
0 0 0 0 421 421
0 0 0 257 0 257
102 0 0 0 0 102
0 0 0 0 2787 2787
138 0 0 0 0 138
1545 2022 3935 2399 5726 15627
-9 Missing or wild
-7 Proxy respondent
1 Higher Degree
2 First Degree
3 Teaching QF
4 Other Higher QF
5 Nursing QF
6 GCE A Levels
7 GCE O Levels or Equiv
8 Commercial QF, No OLevels
9 CSE Grade 2-5,ScotGrade 4-5
10 Apprenticeship
11 Other QF
12 No QF
13 Still At School No QF
Highesteducationalqualification
Total
-9.001.00
Degree2.00
Diploma
3.00 Higherschool orvocational
4.00 Schoollevel orbelow
educ4
Total
DAMES provides online services for data coordination/organisation
Tools for handing variables in social science data
17/MAR/2010 DIR workshop: Handling Social Science Data
21
GESDE – Search and browse supplementary data on occupations; educational qualifications; ethnicity
17/MAR/2010 DIR workshop: Handling Social Science Data
22
23
Conclusions: What we do and what we ought to do
Research writers routinely select single simplifying sub-optimal collinear categorisations of concepts Due to coordinated instructions [e.g. Blossfeld et al. 2006] Due to perceived lack of available alternatives Due to perceived convenience
To make longitudinal comparative analyses more scientific (cumulative & open to cross-examination) we should… Acknowledge and discuss our equivalence strategies Operationalise and deploy various categorisations (measurement
equivalence) and scalings/arithmetic measures (meaning equivalence), and explore their distributional properties
… and keep a replicable trail of all these activities.. …ideally by using GESDE…!!
References Blossfeld, H. P., Mills, M., & Bernardi, F. (Eds.). (2006). Globalization, Uncertainty and Men's Careers: An
International Comparison. Cheltenham: Edward Elgar. Bosveld, K., Connolly, H., Rendall, M. S., & (2006). A guide to comparing 1991 and 2001 Census ethnic group data.
London: Office for National Statistics. Harkness, J. (2008). Comparative survey research: goals and challenges. In E. De Leeuw, J. Hox & D. A. Dillman
(Eds.), International Handbook of Survey Methodology. London: Psychology Press. Harkness, J., van de Vijver, F. J. R., & Mohler, P. P. (Eds.). (2003). Cross-Cultural Survey Methods. New York: Wiley. Hoffmeyer-Zlotnik, J. H. P., & Harkness, J. (Eds.). (2005). Methodological Aspects in Cross-National Research.
Mannheim: GESIS - ZUMA Zentrum fur Umfragen, Methoden und Analysen. Jowell, R., Roberts, C., Fitzgerald, R., & Eva, G. (2007). Measuring Attitudes Cross-Nationally. London: Sage. Lambert, P. S., Prandy, K., & Bottero, W. (2007). By Slow Degrees: Two Centuries of Social Reproduction and
Mobility in Britain. Sociological Research Online, 12(1). Li, Y., & Heath, A. F. (2008). Socio-Economic Position and Political Support of Black and Ethnic Minority Groups in
the United Kingdom, 1972-2005 [computer file]. 2nd Edition. Colchester: UK Data Archive [distributor], SN: 5666. Mohler, P. P., Pennell, B.-E., & Hubbard, F. (2008). Survey Documentation: Toward professional knowledge
management in sample surveys. In E. De Leeuw, J. Hox & D. A. Dillman (Eds.), International Handbook of Survey Methodology (pp. 403-420). Hove: Psychology Press.
Rose, D., & Harrison, E. (Eds.). (2010). Social Class in Europe: An Introduction to the European Socio-economic Classification London: Routledge.
Schneider, S. L. (2010). Nominal comparability is not enough: (In-)Equivalence of construct validity of cross-national measures of educational attainment in the European Social Survey. Research in Social Stratification and Mobility.
Simpson, L., & Akinwale, B. (2006). Quantifying Stablity and Change in Ethnic Group. Manchester: University of Manchester, CCSR Working Paper 2006-05.
Treiman, D. J. (2007). The Legacy of Apartheid: Racial Inequalities in the New South Africa. In A. F. Heath & S. Y. Cheung (Eds.), Unequal Chances: Ethnic Minorities in Western Labour Markets. Oxford: Oxford University Press, for the British Academy.
Treiman, D. J. (2009). Quantitative Data Analysis: Doing Social Research to Test Ideas. New York: Jossey Bass.
Abstract In working with longitudinal social surveys, we sometimes ignore, or at the very least over-
simplify, challenges of measurement comparability in variables over time. Often, our approaches to harmonization involve searching for categories with consistent names over time, frequently achieving such ‘nominal equivalence’ by reducing the level of detail of measures to the so-called lowest common denominator.
In this paper we highlight some well-known, and some less well-known, comparability problems for categorical measures in longitudinal surveys. We focus on UK studies and their measures of occupational position, educational qualifications and ethnicity. These are widely measured and important categorical variables, whether being central analytical variables, or relevant controls in analyses with a different focus. We propose contributions in the form of (i) new online services that can assist in harmonizing measures, which we have generated through a recent ESRC-funded project on ‘data management’ (www.dames.org.uk), and (ii) our own suggestions on achieving effective harmonization of longitudinal variables, which focus upon documentation for replication, the recognition of age-cohort distributional differences, and a general advocacy of scaling categories as part of a strategy of ‘functional equivalence’. These cannot solve every comparability challenge in these areas, but, we argue, are steps in the right direction.
25
Data curation tool (for collecting metadata)
17/MAR/2010 DIR workshop: Handling Social Science Data
26
[Timing issues] Relation between time of survey sweep and exact timing of interview
NCDS(March)/BCS (April), & relative age effect in Engl. (Jul/A) & Scotland (Jan/Feb)
In some instances, categorical thresholds make timing matter more (e.g. b/a exam res)
…9/11 effects seem evident in the BHPS, but vanish after controlling for age
27
0 20 40 60
After 9/11
Just after 9/11
On 9/11
Before 9/11
Source: BHPS 2001 interviews. N =1132; 307; 4063; 13384 by time
Future looks good Age
28
WhB-a
WhO-b
Ind-c
WhB-c
PkB-a
WhO-c
WhI-c
WhB-b
WhO-a
Oth-b
BA-b
BC-b
Chi-bPkB-bBC-a
BC-c
Ind-b
Chi-c
Oth-a
PkB-c
BA-a
WhI-a
Ind-a
WhI-b
Oth-c
BA-cChi-a
I/II
IIIaIVabcV/VIVII/IIIb
UnemployedInactive
-2-1
01
2D
imen
sion
2 (
22.
1%)
-2.5 -2 -1.5 -1 -.5 0 .5Dimension 1 (58.4%)
a = Born in UK; b = Came to UK before 1970; c = came to UK 1970 or laterN=640295 (Data: Li and Heath, 2008)
LFS pooled data for men, 1991-2005Correspondence analysis dimension scores