My first post here. I'm currently doing a course in uni which just goes through basic statistics and research methods. I am, however, having trouble with my final assignment. In this assignment they just given us a dataset of child sexual abuse (CSA) and a bunch of random variables and we have to construct a research question and our own method of analysis. I guess it's really trying to make us apply what we've learnt or something.
Anyway, I've decided to analyse whether gender has a role in the susceptibility of developing mental and behavioural problems and I'm kinda stuck on how to analyse the data to prove or disprove the hypothesis.
What I've started with is just doing t-tests with history of child sexual abuse (dichotomous) and different dependent continuous variables (depression, delinquency, aggressiveness, anxiety).
I also went into using the ANOVA tests for each of them, however, all these tests had insignificant interactions, i.e. gender:rcsa F(1)=0.32, p=0.5, so on so forth. Does this mean that there is no effect of gender? I'm not really sure on how to interpret this on R-commander.
I've also received a lot of other categorical data such as a categorical form of history of childhood sexual abuse, i.e. non-penetrative, penetrative, none, and all categorical forms of deliquency, aggression, depression and anxiety/depression. Should I also use these data? I already have the continuous form of the data, or is this a way to reinforce my findings?
I've decided to analyse whether Child Sexual Abuse (CSA) is related to psycho-social outcomes, so I'll probably use a chi-squared for the dichotomous outcome (i.e. depressed/not depressed etc.). After which, a t-test would be done to compare the means of the continuous measure of the psycho-social outcome.
For any statistically significant association, I decided to use ANOVA to measure the interaction between gender and CSA.
Lastly, to rule out confounders, I'm just gonna pop in random stuff into a logistic regression.
Does this make sense to anyone out there? Or am I doing a lot of redundant stuff?
It helps if you describe variables in the analyses as either independent or dependent. With independent predicting the dependent variable. It is usually standard to run some exploratory bivariate analyses (e.g., chi-square, t-test, ANOVA) in order to understand crude relationships. Then based on this information, build a multivariate prediction model (such as logistic regression, so you are not popping in random variables, they are based on some information).
It is also important to understand when each of the tests are used. It seems like you keep trying to use ANOVA, but do you realize it is used with the same data as the t-test, but when you have greater than 2 categorical groups. You also never describe your study design. Are all of these data collected from a cross-sectional survey or how are they hypothetically collect and for what type periods?