Effect of schooling on menstruation methods

[laxmi Sharma]

Hello,
I am trying to find the effect of schooling on mensuration methods from Indian context.These methods may include use of locally prepared napkins, cloths, sanitary napkins, tampons, etc.My dependent variable is mensuration methods during periods and independent variables are girls's own schooling and her mother's schooling. but the problem is that my data does not have income aspect which can affect schooling of the girl.
So, to control the income, I am using household fixed effect and district fixed effect which includes caste, religion, region, age and living standard. I want to know whether is it okay to consider these effects? Because as far as I know we use fixed effect in panel data. But my data is not panel. the year is include is only 2015.Please help me out in this.
Another doubt i have is that for mensuration methods, I am using 4 dummies.
Dummy =1, if sanitary napkins is used
Dummy =1 if sanitary napkins/ Local napkins is used
Dummy=1, if sanitary/ tampons is used
Dummy= if sanitary/ local napkins/ tampons is used.
I am confused, why my professor has asked me to create these dummy as my data is already in yes/ No format.
Please help me out in clarifying my doubt.

[Carlo Lazzaro]

Iaxmi:
welcome to this forum.
I think you should go -mlogit-. collapsing all the menstruation methods in an unique, 4-level categorical variable, that will be your regressand.
You can use district,caste, religion, region, age and living standard as regressors (take a look at -fvvarlist- to handle these predictors properly).

[laxmi Sharma]

Thank you sir!

[Carlo Lazzaro]

Iaxmi:
please call me Carlo, just like all on (and many more off) this list do. Thanks.

[laxmi Sharma]

Hii Carlo,
I want to combine district code and state code . How can I combine two variables ?

[Phil Bromiley]

There are many ways to do this. You can use egen with group to generate group ids. Alternatively, depending on the range of the actual numbers for districts and states, you can multiply the state code by 1000 or 10000 and then add the district code so the first numbers are a state code and the last are district. The second may require you set the variable type to double. The advantage of the second is that it makes the combinations automatically interpretable.

[laxmi Sharma]

why should we multiply state code by 1000 or 10000 ? In my data, i just have 36 states and 640 districts..
I am using this command . Please advice me because while regressing it with xtreg command , my panel data is coming unbalanced.
egen state_district = group(state distt)
xtset state_district
xtreg san_nap own_edu cast_trib relign place_type age , fe

[Richard Williams]

The multiplying by 1,000 bit is just to get unique ids. So, for example, state 31, district 112 would get coded 31112.

I like the egen approach better anyway. However, the above approach could produce more consistent results than egen if you have multiple data sets and the included states and districts are not always the same.

I don't know why data would become unbalanced if they weren't before, or why you would care. Different areas can have different numbers of cases.

[laxmi Sharma]

Thanks William,
I got what you are trying to say but when did this,
egen state_district = group(state*1000 + distt)
xtset state_district
xtreg san_nap own_edu cast_trib relign place_type age , fe

In my window stata*1000 is not recognized is coming . what do to?
How can I balance my panel variables for fixed effect ?

[Richard Williams]

No. it would be something like

gen id = state * 1000 + distt

You wouldn't use egen with this approach.

You don't have panel data, right? You have groups of people within states/districts. So, you might have 80 people from New York, 54 People from Chicago, 77 from Los Angeles... The Ns do not all need to be the same.

If you are actually running xtreg and getting errors, show your commands and output. Just from what you've said, you've expressed concerns about not being balanced, but I can't see if that concern is based on your results or not.

The term xt is a little misleading, because you can use the commands for things besides panels. For example, you could have a sample of schools with multiple children from each school.

Even if you did have panel data, the panel does not have to be balanced to use fixed effects.

So, unless you're somehow getting an error like "panel must be balanced" or "fe option not allowed" I don't know that you have a problem. And if you do, it might be for some other reason than what you think it is.

Showing us actual code and output will give us a better chance at helping you.

[laxmi Sharma]

Hello Richard,
Now this time, I have used this command.
gen id = state*1000+ distt
xtset id
xtreg san_nap own_edu cast_trib relign place_type age , fe

Also, please have a look on my output . I am not sure whether my fixed effect is correct or not. Although my P value for most of the case is significant.
Please look at the attachment below.

[Richard Williams]

If you've got up to 1,366 observations in a group I suspect you want

gen id = state*10000+ distt

But, it may be that you got unique ids anyway.


I'm not seeing anything obvious that concerns me about the results. Are the signs of the effects plausible? What worries you?

But, is the dep variable, san_nap, continuous? I got the impression from the first post that it was categorical. If so, why are you using xtreg?

I'm confused, because the original question seems to have gotten dropped but I don't see what you actually did with your DV. How is san_nap coded?

[laxmi Sharma]

Actually coefficient of caste_trib relign and other independent variables which I took for fixed effect are coming negative. According to my intuition, they should not be negative.
My dependent variable is in following form.
I have not just renamed the variable. It is already in yes no type. therefore , I don't have to create any dummy variable for sanitary usage.

[Richard Williams]

If it is a dummy variable you should probably use xtlogit. Except I'm not sure how fe will work with groups this large. You may have to settle for re.

I don't know what to tell you about the signs being wrong. Either your intuition is wrong or perhaps a variable is miscoded. Maybe they are coded the opposite of the way you think they are.

[laxmi Sharma]

Thanks Richard.
I got the correct results now because earlier i did not consider the dummy for caste and other factors responsible for district fixed effect.