I apologize for the rudimentary questions. This was not my study, and I am only tasked with analyzing the data.
I am analyzing data on awareness of tobacco health risks before and after an intervention using R. Before the intervention, participants were asked if they knew health risks of tobacco. There were categorized as 0 (did not know) or 1 (did know about health risks). Then, a 5 minute video was shown. After the video, participants were again asked if they knew health risks of tobacco and again categorized as 0 (did not know) or 1 (did know about health risks).
In addition to before and after binary results for each participant, there are additional demographic data available. These data include Age (continuous), School (categorical variable: graduated high school/didn't graduate high school), and Gender (Male/Female) for each participant. There were n=150 randomly selected participants.
Assess the impact of the video on awareness of tobacco outcomes.
My proposed approach
Classify all participants into one of four outcomes.
PRE 0, POST 0: No Improvement
PRE 0, POST 1: Improvement
PRE 1, POST 0: No Improvement
PRE 1, POST 1: No Improvement
Then, use logistic regression on the dichotomous outcome (Improved/No Improvement), while adjusting for Age, School, and Gender. My understanding of logistic regression is that it requires a dichotomous outcome. The (1,0) group would be classified as No Improvement so that the outcome remains binary. (From my research, I briefly considered treating the data as a 2x2 contingency table using McNemar's test for paired data, but it was not clear how to adjust for multiple covariates. I hence chose the above approach with logistic regression).
1) In logistic regression, how do I get an "overall" impact of the intervention? I was intending to use the Intercept value from the model after adjusting for Age, School, and Gender.
2) This approach defines (1, 0) as No Improvement, though in reality it is actually worsening. While this is unusual, I feel it needs to be accounted for. Hence there are technically three results: Improvement (0,1), No Change for both (0,0) and (1,1), and Worsening (1,0). What kind of models can be used for this type of situation?
3) Does it matter that some people previously knew about health risks? In other words, does the model need to acknowledge a difference between (1,1) vs. (0,0) cases? I am not certain whether logistic regression inherently accounts for this or if it matters.
Sorry for the long wall of text. Any help is hugely appreciated, including guidance on other more appropriate models. Thank you.