cross-tabs or non-parametric independent sample t-test

#1
I have conducted a survey to determine if there is a significant difference in the number farmers and non-farmers which agree to a specific farming policy. I have two variables. One variable identifies if the survey respondent is either a farmer or not. The other variable is binary and will record 1 if the respondent agrees and 0 if the respondent will not agree.

I am not sure if I should use cross tabulations but this will state there is an association not a difference. Or would a non-parametric independent sample t-test be a better option because it will confirm whether a statistically significant difference exists between the number of farmers and non-farmers that agree to a specific farming policy. Can you run a non-parametric independent sample t-test with two binary variables?

I hope this is clear. Thanks for taking the time to read my post.
 

rogojel

TS Contributor
#2
Hi,
to me this seems like a textbook example of a chi-squared test ( what you call cross tabulatoon, I suppose). I do not think there is any need to try other methods. Looking at the chi-squared values per cell will give you the infos you need.

good luck!
 

gianmarco

TS Contributor
#3
I also think that chi-q test should be used, since you are dealing with counts (i.e., frequencies). If you find a significant association, you could use Kramer's V (or Pearson's phi) to have a formal 'measure' of the strength of the association. The use of standardized residuals will help you in spotting which cell contribute the most to the rejection of the Null Hypothesis of independence between rows and columns.

Cheers
gm
 
#4
I think what you really want is logistic regression. You have a dependent variable (opinion) and an independent variable (profession) and you want to see the relationship between them. That calls for regression. Since your dependent variable is binary, you want logistic regression.
 

gianmarco

TS Contributor
#5
Hello,
yes, LR could be used. I believe that the preference of one approach over the other depends on the desired goal (modelling vs. testing the independence hypothesis).

gm
 

rogojel

TS Contributor
#7
hi,
I am not sure but it seemed to me that if the predictor is discrete then the LR is actually defaulting to the chi-squared. Could this be right?

regards