Hi Audrey,

You have an interesting question and it helps to have data to test your ideas.

If I understand your proposal correctly, and I may not, your current dependent variable is inconsistent with a probit or logit model because these require a limited dependent variable (e.g., 1 if the doner made a donor, 0 otherwise). Then the analysis would estimate the probability of a donation based on the values of the independent variable(s).

If you like this approach, then consider it; however, I suspect you will find it more interesting, and more useful, to keep the amount of donation (in dollars) as the dependent variable. Some of your independent variables would be dummy variables and their estimated coefficients would be the expected change in the dependent variable. For example, if female fundraiser {FF} = 1 and 0 otherwise, then then let's say the estimated coefficient on FF is +1.23. This means a female fundraiser is estimated to bring an addition $1.23 compared with her non-female counterpart. Moreover, the t-statistic can indicate if the 1.23 is significantly different from zero.

In addition to changing the intercept, like we did in the FF example, we can add an interaction term as another independent variable to test a change in slope as well. This typically involved multiplying FF times another (typically continuous) independent variable, which I did not recognize in your example, so an interaction term may not be applicable if your model.

I hope this helps!

Rob

What you said looks good to me including your model. A comment: three way interactions are usually uncommon. I would look at two way first. This is an explanatory model so you use explanatory methods. See the two references I have attached. ordinary least squares regression should do the trick. Good luck and best wishes, David Booth PS Gelman's book may help too.

https://www.amazon.com/Analysis-Regression-Multilevel-Hierarchical-Models/dp/052168689X/ref=sr_1_fkmr1_1?ie=UTF8&qid=1533083892&sr=8-1-fkmr1&keywords=Gelman+Susan+Hierarchical

Hi Audrey. I agree with Rob that your estimation strategy will largely depend on the type of the dependent variable. From your explanation, it seems like there are zero values in your dependent variable (i.e. for those who did not participate or donate) and there will be strictly positive values (i.e. any amount of donation). If i understood you correctly and this is the case, then maybe one of the censored regressions is a better option like Tobit or craggit. You can use Tobit to estimate conditional effects (i.e. for those who actually donated), and unconditional effects (for the whole sample - the participants and non-participants). You can do this with Probit but your dependent variable (donation size) has to be binary which is clearly not.

If your dependent variable is count (1,2,3,...25) variable then Poisson is a better choice. Ordered regressions (ordered probit or logit are probably not the right choice either as they are mainly for modeling ordinal dependent variables (i.e. dependent variable takes values like excellent, very good, good, not bad, bad, worse).