2
$\begingroup$

We wish do some statistical analysis on the relation between being an entrepreneur and drinking and are wondering if we can consider the data derived from a survey as ordinal in nature. And if we can rank them as such. The possible answers are as follows:

  • No, I have never drunk alcohol
  • No, I no longer drink
  • Yes, but only rarely
  • Yes, I drink alcohol 2 or 3 days per month
  • Yes, I drink alcohol 1 or 2 days per week
  • Yes, I drink alcohol 3 or 4 days per week
  • Yes, I drink alcohol 5 or 6 days per week
  • Yes, I drink alcohol everyday

The yes answers are in our opinion ordinal in nature. The problem is with the 2 no answers. Is it possible to rank the answers as we did here, or is it better to just group them as non-drinkers?

EDIT: Research question: Do entrepreneurs drink more frequently than non-entrepreneurs (wageworkers)?

Improve this question
$\endgroup$
3
  • 2
    $\begingroup$ What is your research question? Whether you group or not will be affected by the question. What is your independent variable? $\endgroup$
    – Behacad
    Commented Apr 3, 2013 at 21:12
  • $\begingroup$ Do entrepreneurs drink more frequently than non-entrepreneurs (wageworkers)? $\endgroup$
    – Troetelbiertje
    Commented Apr 3, 2013 at 21:18
  • 1
    $\begingroup$ The two no answers do contain an ordering (they're ordered on amount ever drunk for example), but that ordering doesn't relate to the research question which seems to be asking about present behavior, not previous behavior. As such, you'd probably group them together for that purpose; then the combined 'no' can be taken as part of the yes ordering. @Behacad: insightful question. $\endgroup$
    – Glen_b
    Commented Apr 3, 2013 at 21:51

2 Answers 2

Reset to default
2
$\begingroup$

There are a couple options.

Certainly you could rank them; given your research question, I would combine the two "No" responses but not the others. Combining categories throws away information. You could then do ordinal logistic regression with "amount of drinking" as the DV and "entrepreneur" as an IV (with other IVs possibly added).

Another alternative is to take advantage of the precision of your categories and turn it into a "days drinking per month":

No, I have never drunk alcohol  = 0
No, I no longer drink  = 0
Yes, but only rarely = 1
Yes, I drink alcohol 2 or 3 days per month = 2.5
Yes, I drink alcohol 1 or 2 days per week = 6
Yes, I drink alcohol 3 or 4 days per week = 15
Yes, I drink alcohol 5 or 6 days per week = 25
Yes, I drink alcohol everyday = 30

or something similar. Then you could do a t-test, or perhaps a count regression. Is this valid? Well, I think it's at least reasonable. You could do sensitivity analysis with different conversion factors.

Lots of variables are not exactly ordinal and not exactly interval either.

Improve this answer
$\endgroup$
3
$\begingroup$

I would combine both "no" categories into one category. Then, your data is clearly ranked. If you like, you could also see after your primarily analysis if entrepreneurs are more likely to be previous drinkers. The issue is that past drinking seems less related to your research question.

If you want, you could even further segregate the categories such as: no drinking, infrequent drinkers (2-3 days per month or less), frequent drinkers (2-4 days a week), very frequent drinkers (5+ days a week). You can easily calculate odds ratios/risk as well and can determine whether entrepreneurs are more likely than non-entrepreneurs to be non-drinkers, infrequent drinkers, frequent drinkers, etc.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.