| Indicator of sexual violence and child marriage [message #1227] |
Fri, 24 January 2014 12:17  |
DHSuserNY
Messages: 6
Registered: January 2014
Location: New York
|
Member |
|
|
I'm using the Women's Individual file to estimate the prevalence of females who have experienced child marriage (first marriage before age 18) and/or sexual violence in various countries by creating a joint indicator of the two variables. Since the age at marriage variable (v511) is asked of all women in the sample and the violence questions (d108, d125, d124, d126) are only asked to one woman per household in the violence module, what weight should I be using for this joint indicator? If I use the domestic violence weight (d005), the estimate will reflect only those in the domestic violence subsample but will this accurately estimate those who were married by age 18? And if I use the regular weight, it seems like it will underestimate those who have experienced violence.
Any advice is appreciated.
Thanks!
|
|
|
|
|
|
| Re: Indicator of sexual violence and child marriage [message #1243 is a reply to message #1228] |
Mon, 27 January 2014 11:05  |
DHSuserNY
Messages: 6
Registered: January 2014
Location: New York
|
Member |
|
|
Great, thanks.. that's helpful!
Another related question, is there a rule of thumb for the number of observations using the DV sub-sample that are considered too small for reliable statistics. For example, with the full sample, DHS usually states under 25 observations is too small to be reliable, and 25-49 should be used with caution when interpreting statistics.
Thanks again!
|
|
|
|
|
|
| Re: Indicator of sexual violence and child marriage [message #1480 is a reply to message #1243] |
Tue, 04 March 2014 14:05 |
Reduced-For(u)m
Messages: 292
Registered: March 2013
|
Senior Member |
|
|
You could always do the power calculations and get some idea for yourself. Stata 13 has a new "power" command, or you could leaf through this paper: http://www.stat.columbia.edu/~gelman/stuff_for_blog/chap20.p df
Or you could just use one of the online power calculators, like this one: http://www.statisticalsolutions.net/pss_calc.php (I just picked this one out of all of them because I played with it once - some offer more alternatives)
Basically, if you have some idea of the standard deviation of your outcome, and how big an effect you want to be able to detect, you can figure out the necessary sample size to get 80% (or whatever) power at 95% (or whatever) confidence pretty easily. I would be that the DHS recommendations come from some calculation like that (or just the old N=40 --> Normal Distribution of Ybar rule of thumb)
For instance, to detect an effect of .5 standard deviations of your outcome with "standard" parameters (alpha = .05; beta = .8) you need 63 observations in each of treatment and control groups (assuming iid random treatment). If you aren't thinking in these kinds of comparisons, the simple power calculations may still give you some idea of a needed sample size - just know that that is an absolute minimum size and should probably be internally revised upward for any hope of finding the effect of interest.
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
