1
$\begingroup$

I have two samples (each nurses) one from 2008 and the other drawn 2013. Both got the same questionaire to fill out. Now I would like to pool the data but do not know if I just can do it. My idea was to check if demographics of the attandents have the same distribution using a two sample KS. I also would perfom a t-test on the item I would like to pool and use for my analysis or maybe test if the standard deviations but do not differ but I do not know if this is enough. It would be great if I could get some advice here.

$\endgroup$
3
  • $\begingroup$ Of course you can pool the data. Are you planning to use a particular software? This can be easily implemented in Stata. I can provide an answer for it. $\endgroup$
    – luchonacho
    Apr 21 '16 at 9:21
  • $\begingroup$ I am using R or SPSS unfortunately I do not use Stata. $\endgroup$
    – ruedi
    Apr 21 '16 at 9:23
  • $\begingroup$ They are beyond my knowledge. But just google two sample KS R or SPSS. I'm sure there are plenty of examples. $\endgroup$
    – luchonacho
    Apr 21 '16 at 9:24
1
+50
$\begingroup$

Whether or not it makes sense to pool your data is as much a question for an expert in your field as it is a statistics question. There are a few questions that you can ask yourself to decide if it makes sense:

  • Is there any reason for you to believe that the two populations are different? For example, were the respondents from very different hospitals (large institutions in the first one, small ones in the second); or from different countries? If the answer is yes, you will need to find ways to account for this difference in your analysis.

  • What type of analysis are you planning to do? If you are doing some regression, controlling for which survey the respondent answered might be enough.

The fact that there are several years between the two survey might need to be accounted for in your analysis. It is possible that between 2008 and 2013, the nurse population has changed a lot. This would lead to differences in the correlation between your survey items between the two groups that are real differences rather than random variation. If this is the case, you need to account for the difference between the group but also ask yourself if it makes sense to treat those two samples as representative of the same population. Even if you decide pooling the data makes sense, it will probably be necessary to account for which survey a respondent answered (in a regression context, that would be a control variable). Gelman and Hill (2007) have a good discussion of pooling in the context of multilevel models (chapter 12, p.252-259), which I think explain neatly the difference in results you can expect from pooling your data with and without a control variable for which group a respondent belongs to.

$\endgroup$

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.