2

I have two independent variables: "vaccination status" and "antibody titer assessment". Vaccination is coded as 0 and 1. The other variable is supposed to explain if the person has checked that whether the vaccination has increased his antibody titer sufficiently. The second variable has two states: yes or no.

The problem is that people who are not vaccinated do not need to check their antibody titer (it is not logical). So I don't know if I should leave the cells in the column pertaining to tier check blank or not, when the persons' vaccination status is zero. Or should I enter zero even for the lack of antibody titer assessment, even if the person has not received any vaccination in the first place?

The difference between the two is that if I fill those blank cells with zero, these two independent variables would become highly correlated (source of multicollinearity). Otherwise, if i leave those cells blank, I would loose data in listwise deletion unless I use other methods such as pairwise deletion or mean substitution (in SPSS).

What do you suggest? Do you have any other ideas that these two options?

gung - Reinstate Monica
  • 132,789
  • 81
  • 357
  • 650
Vic
  • 1,126
  • 5
  • 17
  • 28

1 Answers1

2

You can make them into one independent variable with 3 levels: No Vaccination, vaccination no titer and vaccination titer. This will help in many contexts, e.g. if this is to be an independent variable in a regression, which is what it sounds like from your question.

Peter Flom
  • 94,055
  • 35
  • 143
  • 276
  • 1
    WOW THANKS A LOT dear Peter :) Yes it is Regression and I would do as you mentioned. :) – Vic May 27 '13 at 13:55
  • Update: dear Peter, it was interesting to see that when not combined together, those two independent variables (vaccination and titer assessment) had opposite effects!! So I think I can't combine them into one variable, if I want their effect to appear as statistical significance. When I combine them together as an ordinal variable with 0, 1, and 2 values, the result becomes quite non-significant. Do you have any suggestions for such a situation? (I am currently reporting them independently [and gladly VIFs are not large (VIFs < 1.5). But was curious to know if they can be combined too? – Vic May 28 '13 at 16:33
  • (I mean in a way that their total effect could emerge as statistically significant). Maybe I should code them differently? Or maybe in such a situation I can't combine them. – Vic May 28 '13 at 16:34
  • 3
    You don't change the analysis because the results were not satisfactory. As two IVs, this makes no sense. But is it ordinal or categorical? – Peter Flom May 28 '13 at 19:30
  • No no I didn't mean I want to re-analyze to obtain satisfactory results. What I tried to say was that "the second variable [titer assessment" might not be considered an extension to the first one "vaccination" in an order"... Ah I see, I was treating the combined variable as Ordinal. But the true form is considering it Categorical. I think a categorical variable with more than two levels (llike this combined 3-level variable) should be dummy coded, right? If so, I am guessing maybe a better form is to check those two variables separately (as if I have dummy coded the combined variable back to – Vic May 29 '13 at 06:17
  • its two-level ingredients (I mean vaccination and titer check)). – Vic May 29 '13 at 06:20
  • I hope my interpretation was OK? :) Please kindly notify if you thought I am wrong again. Thanks a lot again for your time dear Peter. – Vic May 29 '13 at 06:21
  • 2
    Yes it should be dummy coded (or some variation on that, such as effect coding). The problem with two variables separately is that one combination can't logically exist, so there is collinearity and confusion, or else a lot of missing data on one variable – Peter Flom May 29 '13 at 10:00
  • Many thanks again and sorry for bothering. Gladly the VIFs didn't get too high (they are about 2 for these two variables and about 1 for most of other ones except a couple of other ones which are about 3). So I think I can treat them separately (thanks to your valuable comments). Thanks for "effect coding". I will check it out too. :) God bless you! :) – Vic May 29 '13 at 20:12