Export thread

It's been long enough since I last had to us stats that I've forgotten a lot.

I'm doing a project. It's supposed to be a proposal for a program evaluation at a women's shelter. One thing we want to show is which services are helping women the most. That makes my dependent variable successful completion of the shelter's program.
They offer 5 stand alone services in the program. The women can choose to participate in any combination of them. So my independent variables could look like this:

A, B, C...
A+B, A+C, and so on...
A+B+C, and so on...
A+B+C+D, and so on...
A+B+C+D+E

We're already (hypothetically) collecting the data from the agency's records, so we can see what services clients participated in and who had a successful outcome. That's going in a table which shows how many successful clients participated in X number of services, where X is 1 - 5.

I really don't want to write out a table listing every possible combination. So if I can say we're going to run *this analysis* to find the combination that works best, it will give my professor the information she's looking for. I'm just not sure what it would be. What analysis do I run to find which variables alone or in conjunction are contributing most to successful outcomes in the program?


I thought at first it should be a correlation, but I think that's what I'd run once I determine which program or combination is leading to success to prove the correlation with a successful outcome.
Now I'm looking at factor analysis or multi-level modeling. I'm pretty sure it can't be an ANOVA since the independent variables aren't in categories like gender or age.

Help?

Wow. If I had some time, I think I might be able to come up with something helpful, but I don't.

Suggestion: break down your overall goal into pieces, determine what you want to measure, and then run a meaningful and appropriate statistical analysis on that. Build it back up toward your main goal.

If that doesn't make sense, blame the lack of sleep on my end. :durr

I recommend a multiple regression analysis. You could do a five way factorial anova but the number of groups is very large in that case. Let's chat in message when I am not on my phone.

MindDetective said:

I recommend a multiple regression analysis.

Click to expand...

That is what I'm looking for! I don't need to actually do the analysis or write anything more than we're using a specific statistical analysis to show which services work best. If you still want to chat, great! Otherwise I have all that I need

Okay, sound good!

jwhouk said:

Wow. If I had some time, I think I might be able to come up with something helpful, but I don't.

Suggestion: break down your overall goal into pieces, determine what you want to measure, and then run a meaningful and appropriate statistical analysis on that. Build it back up toward your main goal.

If that doesn't make sense, blame the lack of sleep on my end. :durr

Click to expand...

It's ok. I couldn't even spell statistics right.

So she says we don't have enough cases in our sample to use a regression. FFS this is a small, church-run agency, not a major non-profit. It was suggested we use a bivariate analysis. I cannot imagine having to run that as a 5-way ANOVA. I'm going to back to just saying what results we're looking for (a specific combination of services that contribute to a successful outcome) instead of exactly what test will get us the result. No one else put the name of a statistical analysis method in their presentation, so it won't matter if we don't have it either.

I would give a hug rating if I wasn't on Tapatalk at the moment.

I hugged you for the thought.

WasabiPoptart said:

So she says we don't have enough cases in our sample to use a regression. FFS this is a small, church-run agency, not a major non-profit. It was suggested we use a bivariate analysis. I cannot imagine having to run that as a 5-way ANOVA. I'm going to back to just saying what results we're looking for (a specific combination of services that contribute to a successful outcome) instead of exactly what test will get us the result. No one else put the name of a statistical analysis method in their presentation, so it won't matter if we don't have it either.

Click to expand...

Yeah, number of cases would be important. That's also true for running a complicated factorial design. Ultimately, the problem is that with that many variables, you need to have a crap load of observations. The solution for a small organization like this would probably be to make it longitudinal so that you could gather a lot more observations over many years. If you are talking about a small-scale study, you shrink the number of conditions/variables to make it more manageable as a bivariate analysis (such as only focus on two interventions).