Are there other variables within your model as well the confounding variables you are interested in?
Hello... I'm in a project to investigate the prevalence of intrusive thoughts in patients diagnosed with major depressive disorder and non-clinical population using chi-square statistics. The presence of intrusive thoughts (in 7 domains) is coded as 1 and the non-presence of these thoughts is coded as 0. I want to investigate the effect of confounding variables (age, obsessive-compulsive symptoms, gender and the level of education) on the prevalence of intrusive thoughts using logistic regression. note that age and obsessive-compulsive symptoms (total OCI-R score) are continuous. How can i investigate the effect of these confounding variables using logistic regression by SPSS?
I want to compare the p-value and odds ratio of the prevalence of intrusive thoughts before and after considering the role of confounding variables. any suggestions?
Are there other variables within your model as well the confounding variables you are interested in?
Hello... maybe other confounding variables can effect the model but I consider only age, gender, level of education and total obsessive-compulsive score.
I am confused what you are doing. First logistic regression is not chi square but you say you are doing both. Second you say you want to investigate the impact of confounding variables, but it looks to me like you are using them as predictor (independent) variables. What are they confounding if they are not predictor variables? Are you trying to control for their impact on the prevalence of thoughts and you have some other predictor they are confounding?
You say you want to- before and after what? Is there some intervention going on you are trying to measure, while controlling for what you call confounding variables? If this is what you are trying to do then you would look at the Wald statistic of this predictor not the overall model value normally.I want to compare the p-value and odds ratio of the prevalence of intrusive thoughts before and after
If your response variable was interval I would suggest repeated measure ANCOVA, but that won't work with a binary response variable.
"Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995
I believe that I follow your description for the most part. You can do this with logistic regression and I would drop the chi-square approach if moving forward with regression.
You have design issues, but if I understand you want to compare the proportion of intrusive thoughts (IT) in your two subsamples. You would run the logistic regression with and without the covariates of interest and see how the odds of IT between the subsamples changes. However if you have not run logistic regression before, your example may be a difficult one for a first attempt. You will want to get a good book on logistic and make sure you cover the assumptions and tests as well as fully understand what you are doing. We are happy to answer specific questions at that time.
Your result will be limited by your design. Issues that jump out will be related to how you select the samples and inclusion/exclusion criteria, variable measures and collection too. Interpretation will be limited to an association but definitely no causative relationship, since we do not know if IT causes compulsion or vice versa, plus a slew of other confusion. Also, you will should draw a causal diagram or path analysis to really understand what or how a confounder may or may not function in your scenario. As Noetsi mentions, it seems like you want to control for these variables, but whether they confound is a complex question.
vahid1990 (09-17-2015)
If you dependent variable is a proportion, and not a binary result, than you can use linear regression. That is a lot easier to use for most and I would (if again the dependent variable is a proportion).
"Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995
Yeah, I may have mis-lead you there. They seem to have a binary outcome, but with two groups of interest I was calling it a proportion (e.g., group 1: 50% have IT; group 2: 85% have IT).
Tweet |