| Suggestion in categorizing preceding birth interval variable (b11) [message #8137] |
Thu, 27 August 2015 14:52  |
suzzon
Messages: 18
Registered: July 2013
Location: Bangladesh
|
Member |
|
|
Hello DHS experts,
I am working on the socio-economic inequality of the infant mortality in Bangladesh using data sets of all the BDHS survey from 1993/94 to 2011. I want to use "preceding birth interval" (variable b11) as one of the determinants in my analysis.
In most of the related literatures, I have found that authors excluded the "no preceding births" cases (the missing values of b11) from their analysis/regression models. I have also found a literature where the "no preceding births" cases are added to the highest category of the preceding birth interval variable claiming that both highest category and no preceding births cases have same "biological" meaning. For example, the categories of preceding birth interval variable are coded as (<2 years), (2-3 years) and (4 or more years + "no preceding births").
Excluding the "no preceding births" cases will surely reduce the sample size and I am trying to avoid it. So my question is, is it ok to add the "no preceding births" (the missing values of b11) to highest category of preceding birth interval variable? Or, is it must to exclude the missing values if I want to use preceding birth interval variable in regression analysis? Or, is there any other way to include them in the analysis. Can you suggest me what is/are the appropriate way/s?
Any ideas would be very welcome.
Thanks in advance.
Moin
Note:According to the discussion in the forum, I have already excluded the children born less than a year before the survey as they have not been fully exposed to the risk of infant mortality.
|
|
|
|
| Re: Suggestion in categorizing preceding birth interval variable (b11) [message #8139 is a reply to message #8137] |
Thu, 27 August 2015 15:37  |
Reduced-For(u)m
Messages: 292
Registered: March 2013
|
Senior Member |
|
|
This is not so much a DHS question as a general methodological one with no specific answer. You seem to have thought through the options, so I think it is up to you to make a decision on your analysis.
One option you didn't mention was that, since you are using categories of preceding birth intervals (and not a polynomial fit using birth interval as a continuous variable) that you could just include a category for "first born". This would buy you some power in your regressions because you'd have more observations that "nail down the X's" - meaning more information for estimating your covariates, even if you don't use that information to estimate the preceding interval point estimates directly. This would be one option that splits the difference between the previous two.
In my very limited experience, the infant health people I talk to treat first births as their own category because they are biologically different than other births. Also, if there is a long preceding birth interval, it means the mother is fairly old, and the family planning decisions are out of the norm, and so you have fundamentally different kinds of women having first-born babies compared to women having after-long-pause babies, in ways that are partly observable but likely partly unobservable. So I'd urge caution in just lumping them in. That said - if it is a norm in the field, then that is how they do it and they probably know more about infant health than I do.
When all else fails - do it every way and report the different estimates in a robustness table. If the two estimates are way different, something weird is going on, and maybe it has to do with the kinds of households who have long-lay-off babies and the kind that have first babies.
|
|
|
|
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
