3
$\begingroup$

First and foremost I would greatly appreciate any help you can provide me with.

I am writing my undergraduate thesis on the rise of populism in France.

The relationship I am trying to better understand is as follows: Can we explain the rise in popularity of the National Front (political party) between 2000 and 2014 through the rising number of immigrant populations immigrating to France?

The data I have so far are the average popularity ratings per year for the political party, and the change in immigrants in France?

$\endgroup$
  • 1
    $\begingroup$ It will be extremely difficult to support your hypothesis in any way. You are claiming a causal relationship between two observed values (and correlation doesn't imply causation). I don't rule out that a convincing approach is possible. But also I don't think a simple OLS will do it. $\endgroup$ – ziggystar Oct 26 '14 at 18:11
  • 2
    $\begingroup$ Short answer: No. To be valid, any sort of statistical analysis would require much more sophisticated approaches (e.g. the realm of causal inference) as well as more data (since you'd need to address the issue of possible confounding variables), etc. $\endgroup$ – Steve S Oct 26 '14 at 18:12
  • $\begingroup$ When these comments were posted I was just writing my own answer. In principle ziggystar and Steve S are saying the same in a much more condensed fashion. My answer tries to offer a potential way around it but in general their points are very important. $\endgroup$ – Andy Oct 26 '14 at 18:24
  • $\begingroup$ OLS assumes the data is IID. Points in time series are dependent on previous points, which can produce a trend over time. If two time series show an upward trend during the same period, they'll be correlated even if there's no causal connection between them. For example, global temperatures were rising during the same period, but we don't look to global warming to explain the National Front's popularity. Tyler Vigen has great examples of time series producing spurious correlations. $\endgroup$ – Lizzie Silver Oct 27 '14 at 19:53
4
$\begingroup$

If you are trying to establish a causal relationship between higher popularity or vote shares of the Front National with immigration then the answer is no. The reason is that you cannot construct a credible counterfactual, i.e. a comparison situation from which you can infer what would have happened to the vote share of the Front National in the absence of higher immigration. At best you can find a correlation between the two variables.

What you are trying to do is very difficult in practice and requires good data and appropriate estimation techniques. Running OLS only will not be sufficient because of omitted variables that affect both immigration and the Front National's vote share. This will bias your results and you may draw wrong conclusions from your data (here that explains the problem of omitted variables). Other econometric/statistical problems include reverse causality: immigration may increase the support of this right wing party but an increase of the support of this party may reduce immigration - in this case immigration is again correlated with the error term, i.e. your estimates will be biased.

In the perfect setting you would have two parallel worlds in which one has a France from 2000-2014 with immigration and the other one would have a France from 2000-2014 without immigration, all else equal. Then you could run plain OLS to uncover the causal effect of immigration of the Front National's vote share - of course we only have one world, unfortunately.

As an alternative: try to find county level data with vote shares for the Front National and identify counties that have immigration and some that don't. You can then compare those counties (with and without immigration) that have similar trends in the vote share for the Front National. Then you can apply a technique called difference in differences (see for example here for an explanation) which has higher hopes to uncover the effect of immigration on the vote share.

I say it has a higher hope because this model does not come assumption-free. You must make sure that you have included other time-varying factors in your difference-in-differences (DiD) regression that affect the vote share like unemployment, the vote share of other parties, etc., that you have two or more counties that are actually comparable in their pre-immigration characteristics, and that no other policies happened at the same time. The problem is that immigration is not a one-time event but immigrants flow over time. Yet DiD would be a good starting point and definitely at a very high standard for an undergraduate thesis.

$\endgroup$
  • $\begingroup$ Hi Andy, Thank you so much for the response. Could you elaborate on pre-immigration characteristics? Does this mean that I would find a region with lower rate of immigration and compare it to a region with a high rate of immigration? Or is this a comparison of the same region over time? $\endgroup$ – DannyGU Oct 26 '14 at 19:08
  • 1
    $\begingroup$ Pre-immigration characteristics are mainly the vote shares of the Front National in the two counties under comparison which need to have the same trends (not necessarily same levels) of Front National vote shares before the immigration occurs. This is what makes the two counties comparable regarding the outcome. You then compare the vote shares in the county without immigration to that county with immigration (which in the absence of immigration should have evolved in the same way as the county without immigration). This is the essence of DiD. $\endgroup$ – Andy Oct 26 '14 at 19:25

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.