I'm trying to conduct an analysis on whether or not there is a significant difference between two populations and different attributes.

Population 1 = 81 observations

Population 2 = 621 observations

For both of these populations I have two variables:

v1 = rating low, below, average, above, high

v2 = score low, below, average, above, high

For each of these variables I have made binary assumptions of 1 = yes, 2 = no such as observation 1 from population one has [0 0 0 1 0] --> v1 = above, and [0 0 0 0 1] --> v2 = high, I have this attribute for all of the 702 observations. I now want to find significant relationships/differences both within each population and between the populations.

For example:

In population 1 there are a significant relationship between v1 = average and v2 = high.

And population 1 has significant higher v1 (high) vs population 2. The problem here is that I have different population sizes and I'm not sure how the handle the case that they aren't normally distributed.

Does anyone have any idea how to construct such a test? Or do I have to few observations to actually be able to say anything about the data at all?

Thanks in advance!

  • $\begingroup$ What do you mean that there is a "significant relationship between v1 = average and v2 = high", eg? Usually, we test to determine if there is a relationship between 2 variables. I'm not sure what it would mean for there to be a relationship b/t 1 level of 1 variable & 1 level of another. $\endgroup$ – gung - Reinstate Monica Apr 5 '16 at 14:40
  • $\begingroup$ @gung Thanks for response, what I mean is that I would like to check whether or not there is a relationship between having an average score in attribute v1 (quality) and attribute v2 (price). I understand I will have to run this for each and everyone of the variablies such that checking the relationship between v1 = low vs all of the v2 = low/below/.../high separately. But I'm not sure of which tests that are most appropriate for this kind of data. $\endgroup$ – Kev1000 Apr 5 '16 at 15:28
  • $\begingroup$ I.e. if there is some sort of relationship, high quality always results in high price, then I would like to check this with significance. @gung $\endgroup$ – Kev1000 Apr 5 '16 at 15:39
  • $\begingroup$ I have read a bit into the subject and understand since the scales v1 & v2 are actually scores from low--> one should use ordinal variables? What would be the difference if i still coded them as binary? I know that there are fewer cars rated as "high"/"low" vs the ratings inbetween these two grades, thus the scale does not have similar lengths of the intervalls. $\endgroup$ – Kev1000 Apr 5 '16 at 18:32

Sorry for the brief response, but I have to run soon and wanted to leave you some information because it has been a while since somebody has responded.

If I were you I would go with the ordinal approach you mentioned in your last comment. Look into the Cochran-Mantel-Hansel class of test statistics, specifically $M^2$ statistic, which tests for independence between two ordinal variables. If the $M^2$ is significant you can go on to look at marginal relationships to gain some insight as to which levels are related to others.

If you have access to it, chapter 3 of Alan Agresti's Categorical Data Analysis book is a good reference for this.


  • $\begingroup$ You're welcome. I hope it helps. $\endgroup$ – D.Mercer Apr 6 '16 at 19:45

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.