which method would be appropriate probit, logit, or anything else?

[ABNGA MANU]

Good morning, everyone,

I am not very familiar with existing econometric methods, especially with survey data. by talking to a friend and he kind of confused my ideas.
Please, I have survey data for a country with 1000 respondents. this survey was carried out in several countries on a T date. (cross-sectional data)
as part of my study I would like to
1) study the influence of a variable X on a variable Y. (X and Y are categorical variables (Yes or No). I therefore think that the simplest would be a probit model...
however,

2) I would also like to see if there are differences in the influence of X on outcome Y depending on the countries. In other words, is it that for a country i, the influence of X on Y is different compared to a country J?
For this reason I would like to study this difference within a set of countries that are grouped under a sub-regional set that includes 8 countries.
if there are differences (which I certainly think there are) I can explain it by cultural, economic or other factors specific to each country.

My question is what method should be used to do this?
8 countries is that enough? knowing that I have a total of 8000 individual observations in my database ( from different countries! 1000 observations per country) and that for theoretical questions I would like to study only this sub-regional set. I have assigned each country a value from 1 to 8 and all other variables are correctly coded and correspond to the questions asked of respondents.

is a random effect probit or logit fix effect appropriate? before verifying by an intuitive haussman test a probit or logit is it suitable for this kind of two-level analysis I want to do?

should countries must be grouped according to shared characteristics and a differential analysis be conducted ( it's not my first idea but wy not )?

Has anyone ever had to do this kind of analysis? Any leads to provide me with?
thank you for your answers

[Carlo Lazzaro]

Abnga:
-svy: logistic- or -svy: logit- could be the way to go.
Moreover, you may want to interact the categorical X with countries in the right-hand side of your regression equation:

i.X##i.coutry

[ABNGA MANU]

Thank you very much Mr. Lazzaro!!!
I think the interaction variable could be a good way to analyze.
have a good day!

Abnga:
-svy: logistic- or -svy: logit- could be the way to go.
Moreover, you may want to interact the categorical X with countries in the right-hand side of your regression equation:

i.X##i.coutry

[Carlo Lazzaro]

You, too!
I take the liberty to remind you the survey feature of your dataset. Hence, the -svy- prefix is mandatory.

[ABNGA MANU]

You, too!
I take the liberty to remind you the survey feature of your dataset. Hence, the -svy- prefix is mandatory.

Thank you Carlo.
I don't have any ID in my database. just starting with country's name, weight, gender age and other variables. but nothing that can identify the person being interviewed

Should I create ID for each observations? it's just one year ( cross sectional data)

[Carlo Lazzaro]

Abnga:
the main issue there seems to be the -svy- prefix, that affects your calculation.
That said, I fail to get how you can perform your regression without a person -id-.

[ABNGA MANU]

Carlo Lazzaro Excuse me, I was wrong. there are indeed person id.

However,please, I have another concern.I am gradually becoming familiar with the survey data, excuse my question if it is...obvious!

the survey methodology indicates that:

In economies where face-to-face surveys are conducted, the first stage of sampling
is the identification of primary sampling units. These units are stratified
by population size, geography, or both, and clustering is achieved through one
or more stages of sampling. Where population information is available, sample
selection is based on probabilities ( Global Findex)


My question is for the syset command, is person id PSU?
I saw that person id is the PSU in case we survey does not involve clustering simple. but if I understood correctly the methodological note above this is not the case for my survey.

Here are some of my variables, the rest are socioeconomic variables, my variable of interest and the control variables
I have no varname that contains identifiers for cluster or household so how could I take the survey feature ? svyset wpid_random [pweight = wgt] ?

Thank you for your answer. You are a great help to me

input str24 economy float pop_adult long wpid_random double wgt
"Afghanistan" 19456060 139880545 1.9367538462406673
"Afghanistan" 19456060 111705622 .3328669818807843
"Afghanistan" 19456060 128866217 1.3389056057152011
"Afghanistan" 19456060 138171768 .2756192037052067
"Afghanistan" 19456060 125156190 .6293038931319532
"Afghanistan" 19456060 209050144 .7731674910391302
"Afghanistan" 19456060 167258084 .48178619304218917

[Carlo Lazzaro]

Abnga:
sorry, but I'm not familiar with survey methodology.
To the best of my knowledge, the guru of this stuff is Steve Samuels, who seems to have decided to take some time off the list.
Who knows whether this methodological issue can make him happy to respond and, at the same time, give us the privilege to read his reply on one of his favourite topics?

[ABNGA MANU]

Carlo Lazzaro thank you very much!! I started to look through his posts which I find very informative....as this one https://www.statalist.org/forums/forum/general-stata-discussion/general/1481269-weighting-without-knowing-psu.
I keep going through them..

thank you for you help!