1

I am currently writing a thesis and struggling with this problem: I have election data for N countries over a time period of 25 years. The elections were (obviously) held in different years. I need to conduct an analysis with country fixed effects and my supervisor told me I should use a method for panel data.

However, the data in my opinion is not a panel dataset, because of the different ts. Can someone give me any input which type of model I should use? Can I just pretend it is panel data?

I looked it up in different standard course books and the definition and examples are always datasets with N observations and T time points, whereas every combination was present.

However in my case I have:

country year
AUT 1990
AUT 1995
AUT 2000
BEL 1991
BEL 1994
BEL 1998
... ...

The observations are all in the same time period but since an election is held every XX years the time points vary from country to country.

I then thought that it would be an unbalanced panel data set, however a dataset is labelled "unbalanced" mostly when some observations are missing and in this case I have most observations missing (since for every i there are only 5-8 t out of the twenty-five t).

For every election I used the share of populist voting as the dependent variable and regressed it on different economic indicators and on a dummy variable whether the election took place 5 years after a financial crises event or not.

x <- lm(share_populist ~ postcrisis_event + GDP + CPI + 
                         factor(country_name_short), data_final)
summary(x)

.

I got a really high coefficient for the dummy variable and found that the other indicators are not significant. Comparing that to prior research in this field it does not make much sense.

Hence this is where my doubts came from.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
clara_____
  • 11
  • 2
  • 1
    You have a time-series for each of $N$ countries. But the exact observation times for each country might be different. That is still a panel dataset, there is no requirement that all the series are observed at the same times! That **might** be a requirement of some specific software, then you need to seek out some different software. But you should tell us exactly the format of each time series, election results can be reported in different ways, and how do you take into account differences between election or party systems? – kjetil b halvorsen Jun 05 '21 at 12:40
  • Okay, that is an surprising answer. I looked it up in some standard econometrics works such as the Woolridge and from there I got, that the definition of panel data is that one needs one observation for every i and every t, but then I probably skipped an important information. I use election results of populist parties in EU countries over a period of time as dependent variable. Some countries have elections every 3 years and some others every 4 years. I then assumed a fixed effects model with country fixed effects and did an OLS regression using the lm function in R with factor(country). – clara_____ Jun 06 '21 at 09:57
  • However, the results I got seem a bit strange to me, which is why I started questioning whether it really is panel data. I gathered that it would probably be an unbalanced panel data set, but again - the definitions I encountered were saying that an unbalanced panel data set was if there were "some" values missing. So I thought, that this does not fit my data either. But maybe I have some major thinking error. To answer your question @kjetilbhalvorsen I did not account for differences between election or party systems, is that necessary to do for my purposes? Thank you for your help! – clara_____ Jun 06 '21 at 10:03
  • Please add new information as an edit to the post, we want posts to be selfsufficient, and few people bother with comments! Also please include detailed information about how you defined your outcome variable, length of the series, and your code used. – kjetil b halvorsen Jun 06 '21 at 14:42
  • Thank you @kjetilbhalvorsen I am new to this, I will update it! – clara_____ Jun 06 '21 at 16:29
  • Why does your model exclude the intercept? – Robert Long Jun 06 '21 at 16:47
  • @RobertLong I got the code from https://www.econometrics-with-r.org/10-3-fixed-effects-regression.html and I did not pay close enough attention to their model. You are right, the intercept should be included. I will change that. – clara_____ Jun 06 '21 at 16:55
  • Ok. So do you have any remaining question(s) ? – Robert Long Jun 06 '21 at 16:58
  • I tried it, and it still yields similar results. Maybe that are just the results - but my main concern is that the function does not work properly because of the data's structure. – clara_____ Jun 06 '21 at 17:05
  • @RobertLong forgot to tag you. – clara_____ Jun 06 '21 at 17:13

0 Answers0