3
$\begingroup$

I am doing a research on foreign direct investment in the EU countries. I came across an article in which the authors assign 4 values to a dummy variable, to be more specific, they assign the value 0 for the years before an event and values of 1, 2 or 3 depending on a country's announced EU accession potential after the event.

I am thinking of doing something similar for my research. I will create a variable taking the value zero before a country’s entrance announcement, one after the announcement and two after the official EU accession.

I have three questions concerning that:

  • Would you say that the authors were right in calling their announcement variable a dummy variable?
  • Do you see any problems with my approach? The other option I came up with is to create separate dummy varibles for a country's entrance announcement and it's actual accession.
  • I might also have an independent variable for trade costs. Would including EU accession in the regression be a problem since there might be some collinearity between the two - trade costs decreasing after entering the EU?
$\endgroup$
  • 1
    $\begingroup$ Dummy variables are set of k binary (indicator) variables representing one categorical variable of k categories. However, in most settings it is suffice or even required (to overcome multicollinearity) to use only k-1 of the set $\endgroup$ – ttnphns Dec 1 '18 at 14:31
  • 1
    $\begingroup$ Dummy variables are by definition dichotomous. Are you sure the article isn't talking about 'one categorical variable with four levels', that is being encoded as 'four dummy variables'. $\endgroup$ – Martijn Weterings Dec 1 '18 at 16:36
1
$\begingroup$

From your description it sounds as though this could be treated as an ordered categorical predictor 0 = before announcement, 1 = after announcement, 2 = process completed. If you want to treat these as ordered and equally spaced so that from 0 to 1 is the same as from 1 to 2 then using a single variable as your predictor is fine. If you cannot make those assumptions then you would need to represent the variable in some other way. You can have two variables representing contrasts between the three situations. They would be dummies; what you have is not.

As for your question about the correlation between this predictor and another: that is not a problem as long as you identify it and think carefully about the implications when interpreting your model.

$\endgroup$
0
$\begingroup$

AFAIK, you can only have 2 values for a Dummy, 1 and 0, otherwise the calculations don't hold. What you can do for three countries is not make a Dummy for country 1 (to avoid perfect multicollinearity) but a separate Dummy for country 2 and 3.

$\endgroup$
0
$\begingroup$

What you should do is to create 2 dummy indicators (not 3 due to what is called the dummy trap http://www.algosome.com/articles/dummy-variable-trap-regression.html) with values 0 and 1. For example entrance would be 0 and a 0 ....announcement would be a 0 and a 1 while acceptance would be a 1 and a 1.

If you made the mistake of using a single dummy and coding 0 or a 1 or a 2 , the one coefficient estimated would reflect a constrained effect where the expected Y is incremented as a multiple of the dummy's regression coefficient or in other words you expect/assume that the change from entrance to announcement is the same as from announcement to acceptance.

$\endgroup$

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.