-1
$\begingroup$

I have two data sets that include variables regarding consumers website usage and click behavior on display ads. The click response per session can only be 1 (clicked) or 0 (not clicked), since only the first reaction within a session was measured. The data are available for several user sessions and would look something like this:

Session_1 Session_2 Session_3 Session_n Impressions_Banner_1 2 4 1 [...] Impressions_Banner_2 1 6 1 [...] Click_Response_Banner_1 1 0 0 [...] Click_Response_Banner_2 0 1 0 [...] Page_Views 2 5 3 [...] Visit_Duration 31 121 60 [...]

I would like to test if it is appropriate to pool the data and combine the variables for banner 1 and banner 2. The question is, how do I test for homogeneity?

The final goal is to run an logistic regression to analyze if/how consumers' website usage behavior has an significant influence on their ad response.

Improve this question
$\endgroup$
3
  • $\begingroup$ Where do the two datasets come from and how are they similar or different? Do they measure usage by the same individual at different time points or sessions, or is there no overlap at all between the observations in both datasets? What do you mean by "homegeneity" and why is it important to test for it for the purpose of your research? $\endgroup$
    – Marquis de Carabas
    Oct 13, 2015 at 20:27
  • $\begingroup$ Actually it is one big data set .Only difference between every session is the click response. If someone clicked on the first banner (click_response_banner_1 =1), the click response for 2nd banner will always be zero. In addition, also both variables can be zero, indicating a sessions with no clicks. Each session illustrates a different visit to the website. It is important because clicks on display ads represent rare events (= problematic for logistic regression). Therefore, I wanna test for homogeneity of the data in order to find out if data pooling is appropriate. $\endgroup$
    – Watermelon
    Oct 13, 2015 at 20:59
  • $\begingroup$ Or in other words: Is it appropriate to merge both click_response and impression variables into one variable each? $\endgroup$
    – Watermelon
    Oct 13, 2015 at 21:07

1 Answer 1

Reset to default
0
$\begingroup$

Yes, if you don't care to predict the difference between the first response or the 2nd.

Make 1 row for each session, and a derived variable that is a 1 for when either one was clicked, a 0 only if none were clicked.

If you wanted to predict them separately, you would just encode it so that your dependent variable was a 0, 1, or a 2 for clicking the 2nd.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Well ok, but do you have any idea how to test statistically if this approach is appropriate? $\endgroup$
    – Watermelon
    Oct 15, 2015 at 11:41
  • $\begingroup$ Not that I know of, this is the setup of your data and the creation of your dependent variable, which always requires some thought and depends on your intent. Both ways are valid but how would you use the predicted score after all is said and done? $\endgroup$
    – Josh
    Oct 15, 2015 at 12:03

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.