Re: dummy variables in child welfare research

["Leona Bull" <l.bull@xxxxxxxxxxxxx>]

Depending on the shape of your data, a repeated measures ANOVA would be
appropriate if you have time 1 (initial screening) and time 2 (discharge)
data measuring the same construct (substance abuse).  It could be that
your data are binary coded (0 for no substance abuse, 1 for any substance
abuse) or it could be coded for degrees or types of substance abuse. 
There are certain statistical assumptions that need to be met for ANOVA
such as the data have to follow normal distribution.  SPSS and other
statistical packages offer procedures to test for this, and any
introductory guide to research methods/ statistics should explain the
background to ANOVA tests.  If your data are unsuitable for ANOVA, you can
use the non-parametric equivalent (the wilcox test) based on change
scores. For change scores to be reliable, the data at time 1 has to be
correlated with the data at time 2 though so run a correlation test first.
 Remember that finding a statistically significant result is not the holy
grail of health services research but rather, a first step.  It just shows
that there was a difference between responses at time 1, and responses at
time 2, over and above that expected by random chance factors alone. If
you find a statistically significant result, you should also look into the
effect size - many statistical packages report this - which is the
magnitude of difference between responses at time 1 and responses at time
2 and therefore some indication of the importance of any intervention
provided between time 1 and time 2.

If the data are anonymous, you can send me the dataset (with codes) and I
will run an analysis for you using SPSS because I am in a very good mood
today :-)

Dr Leona Bull
Research Fellow, Institute of Child Health, UK



> Dear List Members
>
> I am new to the list.  We have data on family substance use and want to
> know if this variable changes over time.  We have 115 families.  They
> answered yes or not at initial screening and then yes or not at discharge.
>  How can we tell if their change was significant over time?  The chi
> square analysis does not seem to be correct.  My boss thinks this is only
> giving us who is more likely to say yes at discharge.
>
> Lisa Canfield
>
>
> ---------------------------------
> Do you Yahoo!?
>  Read only the mail you want - Yahoo! Mail SpamGuard.