0
$\begingroup$

I am doing a survey analysis of a specific country, and I am interested in comparing different regions of that country. Generally speaking (if possible), when a sample is stratified according to region and type of settlement (urban/rural) with the aim of obtaining a sample that is representative on the national level, does it make sense to analyse single regions (or use them as level 2 in a multilevel analysis)? Are the samples representative on the regional level? (According to European Social Survey they are in their surveys, but is this the case also in general?)

Dieter

$\endgroup$
  • $\begingroup$ What is your goal precisely? To compare regions you wouldn't need stratified sampling. There are also fewer gymnasts than football players. Yet if you want to compare their upper body strength, you don't need to adjust sample sizes to reflect population sizes. If however you want to combine the regions into a model representative of the country, then you need stratification. $\endgroup$ – David Ernst Nov 17 '17 at 22:46
  • $\begingroup$ Yes, I see. What I am planning to do is a multilevel analysis, where counties will constitute level 2. My theory says that certain county clusters (macroregions) have in common a certain cultural trait that should have an impact on my dependent variable (DV is on level 1). (We are here talking about a context effect where a level 2 context impacts level 1 behaviour.) By using dummy-variables for these macroregions (on level 1?) I want to find out how much of the level 2-variability can be explained by this context effect. In this case, I guess I will need a stratified sample? $\endgroup$ – Dieter Nov 20 '17 at 10:23
  • $\begingroup$ Yes, either by controlling the sample sizes or by weighting the sample points to artificially adjust. $\endgroup$ – David Ernst Nov 20 '17 at 14:38
0
$\begingroup$

Fundamentally it depends on the design of your survey. Most official surveys are stratified probability samples, where every unit selected in each stratum had a known probability of being selected, and was selected independently of other strata (conditional on the sample design). Under this design-based paradigm, your samples are "representative" (in the sense that the weighted average of the sample is a unbiased estimate of the population mean) on the stratum level, and the national estimate is created by summing the estimates made on the stratum level.

If you aren't using a stratified probability design, then it depends on the design you do have whether it makes sense.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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