I've done a literature review. The variable treat takes only TRUE and FALSE values. It says whether they applied a given treatment in a given article. I want to know if the probability of applying the treatment changes depending on the journal and on the article.type. By the way, I should say that all journals do not have the same frequency of article type.

Here is a data to use as an example:

df = data.frame ( treat=c(T,F,T,T,F,T,T,T,F,T,T,T,T,F,T,T,F,T,F,T,T,T,F,F),

First question:

What statistics test should I use ?

I thought of using GLM (generalized linear model) with a binomial error distribution. Does it seem good to you ?

Second question: Already Answered

summary(glm(treat~journal*article.type, data=df, family='binomial'))

Coefficients: (1 not defined because of singularities)
                         Estimate Std. Error z value Pr(>|z|)
(Intercept)             1.957e+01  4.390e+03   0.004    0.996
journalb               -1.997e+01  4.390e+03  -0.005    0.996
journalc                8.429e-09  8.781e+03   0.000    1.000
article.typeb          -3.913e+01  1.162e+04  -0.003    0.997
article.typec           4.134e-08  1.162e+04   0.000    1.000
journalb:article.typeb  3.884e+01  1.162e+04   0.003    0.997
journalc:article.typeb  3.913e+01  1.756e+04   0.002    0.998
journalb:article.typec         NA         NA      NA       NA
journalc:article.typec -1.916e+01  1.388e+04  -0.001    0.999

When running this, I get several p-values per variable. One p-value for "article.typeb" and one for "article.typec" for example. What does it mean ?

  • $\begingroup$ treat is binary (Boolean) and your response or dependent variable. That being so, your title was the wrong way round. $\endgroup$
    – Nick Cox
    Jul 18 '13 at 13:48
  • $\begingroup$ Remi.b: Your dataset doesn't work: the variable article.type has 27 elements whereas the other two variables have only 24. $\endgroup$ Jul 18 '13 at 14:05
  • $\begingroup$ I'm sorry about the mistakes. I thought I checked my code. I corrected it. Thank you $\endgroup$
    – Remi.b
    Jul 18 '13 at 14:38

Regarding your second question, article is categorical so that R automatically defines several “dummy” variables (other codings are possible), making articles of type A the “reference category”. The coefficient and test for article.typeb represents the difference between articles of type A and articles of type B, the coefficient for article.typec represents the difference between article of type A and articles of type C, same thing for other categorical variables, interaction terms, etc.

  • $\begingroup$ Ok, I got it ! Thanks a lot. I'll wait for an answer to the first question before accepting an answer. $\endgroup$
    – Remi.b
    Jul 18 '13 at 14:43

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.