Need help with MANOVA

#1
Hi there,
I'm working on a research project testing the social control theory of delinquency set out by Hirschi (1969). I have two separate dependent delinquency variables (one for minor and one for serious delinquency) which are standardized scores and three variables which measure social control and have been entered as covariates (two dichotomous variables and one scale). I then also have 8 independent varaiables on such background info as age, gender, level of domestic discord, family structure, etc. I need help constructing a MANOVA as my box test and levene's test both indicate I have violated assumptions. The problem is that my outliers represent truly significant different groups from the norm and and I do not wish to delete them; furthermore, delinquency is not a normally distributed phenomenon and so transforming the variables would be manipulating the data too much. I have already conducted logistic and standard linear regressions to examine the relationship between the three types of variables (independent, intervening, dependent) but now want to take it to the next level of analysis where they are all included in one model which will take into account interactions. Is there any other way to decided which variables can be excluded from the analysis? I am really stuck and my dissertation is due next week!
Thanks,
J
 

JohnM

TS Contributor
#2
It would make sense to put everything into one model to analyze interactions, but you can do that with ANOVA or multiple regression.

Even though you have 2 dependent variables, only use MANOVA if it makes sense to combine the 2 DVs into a linear combination - otherwise, just do ANOVA or mult reg for each DV.
 
#3
Structural Equation Modelling

Hi John,
Thanks for your quick reply. I am still thinking about the best way to go about this final stage of my analysis and have come across structural equation modelling, which Tabacnick and Fiddell say is a good way to test a theory. Since my dissertation is testing a theory of delinquency, would this be a good approach to data analysis. If so, where do my respective variables go (if the social control variables are conceptualized as mediators)?
Thanks,
J
 

JohnM

TS Contributor
#4
I'm not familiar with the Hirschi theory or how it was developed, and SEM is outside of my area of expertise....
 
#5
Hirschi's theory of delinquency states that an individuals bond to social institutions (such as family, school, etc) is directly related to their level of delinquency. Individuals with stronger bonds will tend to be less delinquent due to higher stakes in conformity.
Given the fact, that it is outside your area of expertise, the likelihood of my being able to learn it in the next day or two is probably unrealistic, haha. I want to test Hirschi's theory and my advisor recommended using the MANOVA, but then I ran into the problems I told you about in my first posting. Any other advice?
 

JohnM

TS Contributor
#6
Do separate ANOVAs for each DV, unless your advisor has a strong reason for combining the DVs into a single composite variable.