2

In my study, I have two groups and compared them in terms of blood stem cell levels, which was not distributed normally. For adjusting for age and gender when comparing groups, it is suggested to use regression or ANCOVA in this forum. But my statistician has done it by using binary logistic regression and given me the following tables, and told me that he used the group (Study group:0, Control group:1) as the dependent variable. I am confused about this. Is it correct to use binary logistic regression? My dependent variables are blood stem cell levels (CD34 count and % of CD45) which are not normally distributed. There are two groups, a control and a study group. There was a statistically significance between groups (Mann Whitney U test, p< .001). But I need to compare groups adjusting age and gender.

binary logistic regression model with enter method.

Scortchi - Reinstate Monica
  • 27,560
  • 8
  • 81
  • 248
karaca
  • 43
  • 6
  • Welcome to Cross Validated! Please explain what the statistician did in a little more detail, else we'd be just guessing. – Scortchi - Reinstate Monica Jun 20 '17 at 11:01
  • @ Scortchi, I added the table which statistician made. – karaca Jun 20 '17 at 11:31
  • 1
    Thanks, but that still doesn't explain what they *did*. Is what you called the dependent variable given as "CD34 (+) HCs count" in the upper table, by any chance? – Scortchi - Reinstate Monica Jun 20 '17 at 11:58
  • @ Scortchi, I explained the dependent variable in more detail. – karaca Jun 20 '17 at 12:11
  • 1
    This remains impenetrable. Did the statistician use the 2 groups as the dependent variable? There is no way to tell what was done here or whether it was valid. – gung - Reinstate Monica Jun 20 '17 at 12:56
  • @gung, I asked again to my statistician and he told me that he used groups as dependent variable in binary logistic regression. In this situation, is it correct? – karaca Jun 20 '17 at 13:01
  • @karaca: Please edit the question to include that information. (I think it's gradually becoming clearer.) So there were two dependent variables originally, in two separate ANCOVAs? – Scortchi - Reinstate Monica Jun 20 '17 at 13:10
  • Thanks! It may also help to describe the purpose of the models. There is a similar question [here](https://stats.stackexchange.com/q/41320/17230). – Scortchi - Reinstate Monica Jun 20 '17 at 13:40
  • Indeed this regression model allows controlling for the effects of age and gender but I don't find it straightforward. A presumable easiest way to perform this comparison would have been to directly analyse the CD34 variables as dependent variable (rather than independent variable) and to specify group, age and gender as IVs. The challenge will be to deal with the non-normality of your CD34 variable. – Nicolas K Jun 20 '17 at 21:24
  • @Umka, Yes, you are right, non-normality of my data is a big challange for me. My istatiscal knowledge is very limited. When I search , I found advanced statistical tests like Quade's test, etc. What is your suggestion? This model is acceptable for a journal submission? – karaca Jun 21 '17 at 15:21

0 Answers0