Regression Analyses on SoC Index - Tips? SOC Research and Measurement Discuss Regression Analyses on SoC Index - Tips? in the forums; Good evening! I have used the Sense of Community Index in one of my youth surveys about online behavior. I have data for 295 youth, compiled the index, and I ...

[eparadis]

Good evening!

I have used the Sense of Community Index in one of my youth surveys about online behavior. I have data for 295 youth, compiled the index, and I am now at the stage of running my analyses.

The problem I am facing is that my results are far from normally distributed. I have tried to square the results, cube them, raise them to the power of 2, used an exponential, used the root... and none of these functions have helped normalize the results. Do you have experience analyzing SoC results using regression? Were your results normal? Did you have to use polynomials? Have you ever tried to dichotomize the results and run logistic regressions?
Right now, the data is giving me a headache... and I really wanted to
investigate SoC applied to online communities and youth.

Thank you!

Elise

[dchavis]

Hey Elise;

Hope this eases your headache. Why would you expect the results to be in normal distribution? If you select a community (e.g. internet) that people (such as youth) feel connected to, then the results will be skewed toward the stronger. How connected youth feel to their parents' bridge club may be skewed the other way. You have a high mean (stronger SoC) among members of the community on average, but the distribution than drops off as it gets to a value of 12. That looks You can surely do regression methods legitimately using a sample with a distribution like this. A bigger problem are outliers, which you don't seem to have. Or the extreme frequencies at either end or a"J curve" which often indicates social facilitation (pressure).If you are feeling compelled or being compelled to create a normal distribution, then changing the SCI value to a z-score, is probably the way. SPSS, SAS and other statistical software do it easily.

Your results are normal, the standard of strict replication of a normal distribution is very "old school".

Hope this helps

Tell us more about your study what are your independent variables?

David

[eparadis]

Thanks!

I have run a first set of binary logistic regressions on a first dependent variable, whether or not youth are part of an online LGBT community, however they defined it.

The most important predictors were, in order:

1. If they meet most of their LGBT friends online (vs. equally online and offline, or offline mostly);
2. Whether they are part of an offline LGBT community;
3. If one of their main reasons to go online is to meet other LGBT people.

Other significant predictors were: number of hours spent online, and frequency at which they are likely to meet someone offline that they first met online (negative effect).

Individual-level variables such as race, gender, family tension, college attendance (vs. high-school attendance), whether their parents are aware of their LGBT status were not significant predictors of youth's online community participation.

My SoC analyses will be done only on the subsample who agreed they are part of an online LGBT community (n=295, vs. 476). I expect similar variables to matter, although I believe individual-level variables are more likely to matter in defining the extent to which youth feel a more or less intense community connection. I'll keep you posted!

Ouf! Quite a marathon writing!


Thanks for your support! Write if you'd like to see the full results. I'll gladly send copies.

Elise