1
$\begingroup$

I'm working with European Social Survey data and have seven dichotomous questions related to people´s political participation.

I would like to create a new variable summing them up to be used for regression analysis, but as far as I know, a sum of variables requires at least ordinal data.

Is there a way around that?

$\endgroup$
  • 2
    $\begingroup$ There is no "rule" requiring ordinality for summation. If you want to add these 0,1 features up, to create an overall count, you should feel free to do so. $\endgroup$ – Mike Hunter Mar 26 '17 at 10:48
  • $\begingroup$ On the contrary, it is unusual to sum ordinal scales with 3+ grades (unless you pretend to take them as interval ones), but it is ok in many instances to sum binary scales if they are the same responses (such as 1=selected, 0=not selected; i.e binary ordinal scale). $\endgroup$ – ttnphns Dec 1 '18 at 18:05
2
$\begingroup$

I would be careful about doing aggregation without paying particular attention to what the questions mean. Are some questions more relevant than others, for instance, and are therefore better measures of participation? You might muddy this information by summarizing in this way.

Why not leave them as they are and let the model decide how they should be combined? Also, are you using a linear model? If so, then the summation is only forcing each variable to have the same coefficient, and this is probably not a good idea.

$\endgroup$
1
$\begingroup$

I agree with DJohnson.

It's not uncommon to sum these variables and create a composite measure. I am assuming these measures have some conceptual similarity and make them valid for a comparison. You could also run a Cronbach reliability test on them to see if, at least internally, they are consistent, too.

See page 102, and other topics here for more info.

$\endgroup$
  • $\begingroup$ Many thanks for the advice and the link! That´s great! All the seven variables measure political activity and bear lots of similarities. I already run a Cronbach´s alpha test but it produced a low result (0.54). $\endgroup$ – Minna Mar 28 '17 at 15:18

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.