 |
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: RE: re:linear regression with robust variance estimation
Shehzad-
Without a control group, you will have a difficult time drawing
causal inferences.
Parity after treatment = Parity before treatment +1. Parity is
independently associated with parity. Essentially, the analysis will
depend on comparing the within-woman change in birth weight to
between-woman differences. (I assume the real parity categories 1,2,3
are something like: 0, 1, 2+.)
A few suggestions:
1. Find external control information about expected parity-related
changes in birthweight in the absence of treatment.
2. Account for as much of the between-woman difference as possible,
by adding more predictors to the model. There are many, as any book
on perinatal epidemiology will show.
3. Gestational age is the single strongest predictor of birth
weight. Therefore, four gestational age categories are might not
constitute sufficient control. I would use exact gestational age,
with a spline or fractional polynomial to capture the nonlinearity.
With such a strong predictor, interactions are also possible.
4. However, gestational age and other biological predictors, such pre-
pregnancy weight, might be affected by treatment. Try analyses
without such predictors to get the unmediated impact of treatment .
Mediation analysis has been a recent topic on the list, and I have
found Patrick Shrout's presentation to be very informative (http://
www.psych.nyu.edu/couples/Shrout/APS2006E.ppt). Shrout does assume
the presence of a control group.
I like Nick's suggestion of looking at the second - first difference,
but I suggest that you do this by initial parity category.
-Steve
On Feb 6, 2009, at 9:28 AM, Nick Cox wrote:
I don't know how much help this would be statistically, or
scientifically, but in such cases I would always have a quick look at
second baby - first baby
perhaps plotted against their mean. If babies are better logged or
otherwise transformed, that's a further wrinkle.
Nick
[email protected]
Kit Baum
Ali said
I have a small data set of 136 infants each 2 of them have the same
mother. All mothers have a disease which I am at its effect on
birthweight. These women were diagnosed and treated after giving birth
for the first child. I am trying to measure the difference in mean
birthweight between infants who were born before treatment and those
who were born after treatment. At the same time I want to account for
parity (3 categories 1, 2 or 3) and gestational age (4 categories).
I am using regress with robust option to account for the fact that the
data is not independent. Is there a way to include the mothers
personal number in the model? Or does robust take into this problem
into account?
The model I used is x: regress birthweight i.disease i.parity
i.gestationalage, robust
why not cluster(motherid) ? That subsumes robust and allows for error
correlation among children of the same mother.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/