Mothers Education on children's health [message #9190] |
Sat, 20 February 2016 10:15 ![Go to next message Go to next message](/theme/default/images/down.png) |
simotom
Messages: 4 Registered: February 2016
|
Member |
|
|
Hi
I'm running a linear regression analysis to calculate the impact of mothers'education on children malnutrition in Zambia. I choose to use the KR recode file for the analysis because it has all the data I need about parents education and children health. I have a lot of independent variables and I choose Child Height for Age SD (New WHO, HW70) as the dependent variable. I recoded some of the variables so that 0 would have been the case with lack of something (no education, no toilet facility, bad type of cooking fuel...) and 1 to better situations. However the R^2 and the adjusted one are very little significative. May be the way I recoded the variable? I also tried to recode in several other ways such us giving value between 1 and 5 to all the modalities of the variables, but still the results are not even sufficient to demonstrate something.
I know this is a general statistical problem but I hope someone could help me.
Thank you very much,
Simo
|
|
|
Re: Mothers Education on children's health [message #9210 is a reply to message #9190] |
Wed, 24 February 2016 21:18 ![Go to previous message Go to previous message](/theme/default/images/up.png) ![Go to next message Go to next message](/theme/default/images/down.png) |
Reduced-For(u)m
Messages: 292 Registered: March 2013
|
Senior Member |
|
|
Simo,
In general, the r2 in these kinds of regressions is very small. That is not, in and of itself, any cause for concern. At least not if you are just interested in the effects of particular variables on child HAZ. What is relevant is the t-stat/p-value/std-error on the coefficients of interest.
The r2 is just a bad metric of a "good model" in this instance. Sure - you could up the R2 by, say, adding dummy variables for age-in-months, but really, that will be orthogonal to everything you care about and so will not affect the estimated variability (t-stat/std-errror) of your regression coefficients.
This all assumes that you are just trying to accurately estimate conditional correlations. If you are trying to do something else, you might care about r2, but you'd have to explain what you were hoping to accomplish. If you look at the r2 in most published economics research on this (for instance) they are all very, very low, but it isn't really the thing to be worried about.
Help?
|
|
|
|
|