Actually what i want to do is to build regression model with 4 predictors and 1 dependent variable. But before i have to perform some other work. According with my supervisor one of the steps is - Univariate Logistic Regression analysis
I don't reccomend stepwise (which has a variety of problems and is often criticized in the literature). I assume those terms means pretty much what the do in linear regression. Forward adds variables one at a time until some criteria is met. Backward starts with all predictors than removes them until a criteria is reached. The documentation I posted above should show you how.
The problem is that i cant use linear regression because of the nature of my variables. And i cant find stepwise regression function in binary logistic regression. That.'s why i am a bit confused. Thank u for the link!
Thank u all for your guiding and help. I think i have found what i need --http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/index.jsp?topic=%2Fcom.ibm.spss.statistics.help%2Falg_nomreg_stepwise.htm
but to be honest, right now, i don't understand how to apply it in SPSS.
please can you help me understand the comments of a review...?
I have 1 dependent categorical variable and 4 categorical independent variables (3 dichotomous and 1 with 4 parameters) and i am trying to show that 1 (or more...) out of the 4 independent represents significant risk factor(s) for the dependent var. So, firstly i run correlations with chi square for all the variables and then i proceeded to a binary logistic regression with the same vars. All of the aforementioned to SPSS 20.0. Even i have my results (only one independent was statistically significant eventually) when my paper was reviewed i got back the comment below:
'' I wonder the variables included in the univariate analysis and their results. The multivariate analysis should be performed after the univariate analysis to eliminate the confounding factors. Giving all variables including univariate analysis and the multivariate analysis clearly and the results of the analysis (univariate and multivariate) with OR and CI as a table would be better.''
I didn't know that a univariate analysis is obligatory before proceeding in a binary logistic regression (that's the reason i proceeded only to chi square correlation)...and the truth is that i do not even know how to do it. I mean, all of my variables are categorical so i suppose that i can not use either linear regression or ANOVA...
Univariate analysis implies that you have a single dependent variable (multivariate, in contrast, assumes >=2 outcome variables). Based on your description, your analysis is univariate -- given a single binary outcome.
It seems to me that the reviewer implies a case, in which you include regressors in your model in a step-wise fashion -- i.e., add them one by one and report results for each model in a separate column. For example:
M1: y = x1
M2: y = x1 + x2
M3: y = x1 + x2 + x3
M4: y = x1 + x2 + x3 + x4
I hope this helps.
On a side note, although logistic regression is generally preferable for a dichotomous outcome, a linear probability model (LPM) can also provide consistent estimates. Perhaps, you can use it to ensure robustness of your estimates.