st: ?Test? command for continuous variables?

[omer@demog.berkeley.edu]

Hello, STATA listservers. I was hoping you could help me out with a
statistical issue I have that I?m trying to resolve before sending a
revised paper to a social science journal. Thanks so much in advance for
your help. In one part of my analysis my dependent variables are levels of
various biomarkers (cortisol, DHEAS, dopamine, and others) analyzed
individually. My independent variables of interest are two measures of
subjective social status. Each measure asks respondents to rate their
social status on a ladder that has 10 rungs. One of the ladders asks
respondents to rate themselves in terms of where they think they stand
nationally (taking into account their level of education, income, and job
type), and the other asks them to rate themselves where they think they
stand relative to their community (taking into account whatever factors
they feel are important to them).

In Model A I include just the SES ladder as the key independent variable
with a number of baseline controls (Model 1). I then keep the SES ladder
in the model but add objective measures of status (i.e. educational levels
and occupation prestige score) (Model 2). The idea here is to see whether
the subjective SES ladder predicts riskier biomarker profiles net of
objective indicators of status. In Model B I then do the same set of
analyses (Model 1 and 2), but using only the subjective community ladder.
In Model C I include both status ladders in Model 1 and 2. In Model C I
also perform an F-test using Stata (i.e. after the regression I type ?test
subjective_SES_ladder subjective_community_ladder?). One of my big
questions is whether using the ?test? command in Stata is appropriate
here. I have read that it is only appropriate with categorical variables,
but I?m not sure if this is the case. If it is only appropriate for
categorical variables, why does Stata even produce results for these two
continuous variables?

Another issue has to do with the correlation of the ladders. The two
ladder measures are highly correlated (with a Pearson correlation, equal
to the Spearman rank correlation, of 0.78), but respondents ranked their
social position within their community an average of 0.4-rung higher than
they ranked their position within Taiwan. One reviewer wondered why we
present results from the two ladders since analysis of both ladders ?are
far from independent analyses.? My thinking was that even if the ladders
are highly correlated, they may very well tap very different phenomenon. I
also thought that you never know if one will turn out to be more linked to
riskier biomarkers profiles than other. I also thought that maybe in a
test of joint significance both ladders would together be significant even
if each on their own wasn?t. Does this thinking make sense?

If this does not make sense, one option is to present results from one
ladder and say that the results for either ladder were nearly the same. If
I use this approach, results stemming from which ladder should I present?
Perhaps I should present results for the community ladder because
respondents ranked themselves higher in terms of the community, suggesting
that their standing in the community is more relevant to them. Also, I
sort of like the community ladder question better because it has no
prompts.

Thanks,
Omer

------------------------------------------------------------

Omer Gersten, PhD


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/