Weighting Eurobarometer dataset with STATA

[Ivan Nardini]

Dear Statalist,

I'm a beginner user of Stata and I am conducting a research about religious denominations among UE-28 using Eurobarometer data.

In the Eurobarometer's documentation, there are only codes to weight countries in SPSS.

Can someone know command to set the database to perform a comparative analysis between countries?

I apologise if this question might be a little bit elementary.

Thanks in advance

[Carole J. Wilson]

Ivan,
There are a lot of different Eurobarometer weights. See http://www.gesis.org/eurobarometer-d...ting-overview/. You need to decide which weights you want to use. Some of the weights adjust for population size. In those cases, you need to decide, based on your objectives, if you want to weight by population size (so that a big country has a bigger impact on the results than a smaller country).

[Ivan Nardini]

Hi Carole,

Thank you for your reply.

I read the documentation. The idea would be to build a table that has as i = "religious denominations" and how j = "EU28 and the individual 28 countries of the 'European Union".

So, in your opinion, what would be the best procedure of weighting to obtain such output? and do you know how can I implement it?

Thanks

[Nick Cox]

Cross-posted at http://stackoverflow.com/questions/3...set-with-stata

Please note our cross-posting policy, which is that you should tell us about it. http://www.statalist.org/forums/help#crossposting

[Maarten Buis]

Different weights answer different questions. So none of the weights are universally best or worse; it depends on the exact question you want to answer. So the fact that you want to create a table with rows and columns is not enough information for us to answer your question. We need to know what do you want to do with that table, and what the cells are supposed to represent.

[Ivan Nardini]

Hi Maarten,

I know that it is not the policy of this blog post pictures but in this case "a picture is worth a thousand words":

The final output should be something like that. In this case, I consider european denomination (in aggregate). In my case, I'm going to generalize and draw a similar graph for european region (North, South, Est, West). About your last question, I suppose that cells will represent the number of individuals in a j country who believe i religion.

[Nick Cox]

Difficulty in distinguishing red and green is among the most common difficulties many sighted people have. Putting red and green next to each is thus ill-advised. Using similar colours for atheists and Muslims is especially ill-advised.

Some of the categories here are a little puzzling, but I'll guess that's beyond your control.

What is under your control is how to show such data.

I'd just use a bar chart here. The pie chart should be left in grade school. The multiple colours appear meaningless as well as distracting and the design requires readers to go back and forth repeatedly between legend and graph.

clear 
input id str42 text percent 
1 "Catholic"     48 
2 "Protestant"   12 
3 "Orthodox"      8  
4 "other Christian" 4  
5 "non-believer/agnostic"   16 
6 "atheist"     7 
7 "Muslim"      2 
8 "other religion/none stated"  3
end 

* labmask needs to be installed from Stata Journal files 
labmask id, values(text)

set scheme s1color
graph hbar (asis) percent, over(id) bar(1, bfcolor(none)) blabel(total, pos(base) size(small)) ytitle(percent) ysc(r(-2 50) off)

[Nick Cox]

The grid in the graph above (at 0(10)50) should go.

[Ivan Nardini]

Thanks Nick, you made my idea more clear.

But there remains the problem of what weight i have to use in the database in order to obtain an accurate representation.

I hope that Maarten will provide some useful hint.

[Nick Cox]

I am not Maarten Buis , but I know him very well, and I think he has already signalled clearly that he can't possibly tell you what to do. You have yet to answer his question in #5.

[Carole J. Wilson]

At a minimum, you should include a design weight. Your current numbers describe your sample. If you want your figures to describe "the EU" as a whole, then use the Weight Extrapolated. One observation of a man in Paris, age 30, might represent 20 such men; whereas on 65 year old woman in Rome might represent 30 such women.

Remember to use svy: commands in Stata.

[Ivan Nardini]

Thanks Carol, I'm going to try.