1

I work on Eurobarometer data. There are weights for countries but some countries are divided in parts with specific weights to reconstitute the whole. I need to apply the latter first and the former afterwards. I do not know how to do that.

Example: weights for counties; specific weights for east and west Germany. I have to apply first east and west Germany weights to get Germany and compare it to the other countries with the country-weight.

I am using Spss for data manipulation and Stata for the rest.

mpiktas
  • 35,099
Pier
  • 41

2 Answers2

0

According to the Eurobarometer documentation, the across-country weight already includes the within-country weighting (as is the norm for such data), so you should only need to apply the across-country weighting variable.

It is easy enough to check if this is the case. Just filter the data to look at one country (e.g., Germany), and check that the two weights have a correlation of 1. If this is the case, just use the across-country weight. If no correlation can be computed, because there is no variation in the across-country weight, the solution is to divided the within country weight by its total in each country and multiply by the sample size, and then multiply by the across-country weight. If some other correlation is computed, you will need to get in touch with the people that created the weight.

Tim
  • 3,401
  • Thank you very much for your answer. I contacted the team and they told me both weights should be applied. I have looked into the correlation between within-country weight and across-country weight, correlation is 0.834; so not 1. I will do as you recommend and let you know if it works. Thank you so much. – Pier Sep 17 '15 at 12:18
  • so I did as you said and I obtain weights that are unlikely. They revolve around the 0.000ish. The formula you gave sounds sensible though. What could have gone wrong? – Pier Sep 17 '15 at 13:08
  • 1
    Sorry @Pier, I forgot to write the need to multiply by the sample size. But, assuming that the correlation has only been computed with the data for a single country, I would be a bit nervous about the correlation of 0.834, as this suggests that both weights are trying adjusting for similar things, so their product is likely to be double-counting. I would strongly suggest you pursue this with the people that created the data set. – Tim Sep 17 '15 at 13:36
  • I'm not so surprised about the high correlation in the sense that yes, they adjust for similar things: within-country (WC) aims to allow comparison between East and West Germany; across-country (AC) puts East and West germany amongst the EU countries. The thing is that if I apply your formula: (WCweight / total WC)sample size ACweight; I get figures that do not fit the whole dataset.. – Pier Sep 17 '15 at 14:15
  • 1
    Of course not. If the two weights are adjusting for the same things, it means that it is completely inappropriate to combine them, as you are just double-counting. – Tim Sep 19 '15 at 22:35
  • Thnak you Tim, you're right actually the solution was to substitute the one with the other... Thank you so much for your help – Pier Sep 21 '15 at 08:05
  • Happy to help. Don't forget to mark the answer as Useful. – Tim Sep 26 '15 at 22:26
0

For all those that are working on Eurobarometer, please note that there exists a patch to unite East and West Germany as well as Northern Ireland and Great Britain. The patch can be found on http://www.gesis.org/eurobarometer-data-service/survey-series/standard-special-eb/weighting-overview/

Look at the column on the right. It basically substitutes one weighting with an other for those two cases.

Hope it will help,

Pier
  • 41