1
$\begingroup$

I got my article manuscript back from review and one notion from a reviewer was that in my analyses "[s]tandard errors are not corrected for heterogeneity or intra-group correlation", s/he apparently thinks they should. I'm slightly confused about this.

I have used World Values Survey (WVS) data, where unit of observation is one interviewed person. I have analysed respondents from ten countries of the data (I have country as control variable (coded as a set of dummy variables)), togehter there are some 10 000 respondents in my analyses. Based on what WVS states, the data is representative of all people in the age 18 and older residing within private households in each country. The sampling method of the data is (depending on country) full probability or a combination of probability and stratified (see this, for more specific description by country, select one country here and see Sampling & Methodology). Respondents are interviewed separately.

I have used binary logistic regression and OLS regression. My dependent variables are participation in protests and interest in politics and independent variables are such as education, employment status, income, age, gender etc.

Based on what I have found (for example this and asuming that the term "intra-group correlation" is synonym to "intra-class correlation" (as I would understand based for example on how these terms are used here)), in my case intra-group correlation would occur if respondents were taken from certain groups (they were family members for example) and respondents within these groups had tended to answer similarly. However, this is not the case, the observations should be independent. With heterogeneity I'm equally puzzled.

My question is twofold (or actually fourfold): 1) Is heterogeneity and 2) intra-group correlation something I should be worried about when working with WVS? And if so, 3) How to correct for heterogeneity or 4) intra-group correlation using SPSS?

I am thankful for all advice.

Improve this question
$\endgroup$
2
  • $\begingroup$ Nobody can help you except people who understand the format of the world values survey. Can yu edit to tell us what it is and what it contains? $\endgroup$
    – mdewey
    Commented Jan 22, 2019 at 10:03
  • $\begingroup$ Thank you @mdewey for the helpful comment. I tried to answer it when I edited the post. Is there something else what would be good to write about the world values survey? $\endgroup$
    – Lost in regression
    Commented Jan 24, 2019 at 14:28

2 Answers 2

Reset to default
0
$\begingroup$

SPSS offers several procedures that can address correlated data like these. The CS procedures for complex sample data do this. The GENLIN procedure offers GEE (generalized estimating equations), and the MIXED and GENLINMIXED procedures offer mixed models.

Improve this answer
$\endgroup$
0
$\begingroup$

You asked about this several years ago, so I suppose this isn't relevant to you anymore, but in case others have a similar question, here's a brief answer. As far as I know, the WVS doesn't impose a single, unified sampling strategy. I followed your links and looked at the sampling methodology for the 2014 (round 6) WVS conducted in Jordan. (Unfortunately, can't figure out how to link it directly.) This was a stratified, cluster sample. I suspect this is common for WVS.

Generally speaking, when you have any sort of cluster sample, the observations drawn from the same cluster are not independent, and standard errors that don't take this dependence into account are too narrow. For more on this see my response here.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.