How to Detect Leverage Points/ Influential Observations

[Girlan Oliveira]

I'm working with human opportunity index (hoi) and I am applying the 'hoi' command that is defined by a program in the ssc. In general, it initially estimated logistic regression and using the probability of a positive outcome to calculate the index (HOI).
My database is composed of variables originating from a search of a Brazilian official statistics institute and variables that I created from them using mathematical operations such as average, difference and logarithm. The base contains observations for 27 states over 10 years, and from it I generate the index (HOI) for each state and year, ie, are 270 indices calculated, which is done through a unique code that was created from a request for help made in a previous topic I created this forum (code below).

capture postutil clear
postfile handle ufs ano hoi_1 d_1 p_1 using hoi_sanea, replace

local predictors sexo raca presmae area logrenpcdef nmorad refsexo ncrian medescresp difescresp

local outcomes sanea

levelsof uf, local(ufs)
foreach c of local ufs {
levelsof ano if uf == `c', local(anos)
foreach y of local anos {
display "uf = `c', ano = `y'"
foreach o of varlist `outcomes' {
capture noisily hoi `o' `predictors' [fw = pesopes] ///
if uf == `c' & ano == `y', format(%9.3f) estimates decomp1
if c(rc) == 2000 { // hoi FAILED DUE TO NO OBSERVATIONS
display "Nao ha observacoes ou o outcome nao e dicotomico `o': analise ignorada"
}
else if c(rc) != 0 { // SOME OTHER ERROR AROSE ATTEMPTING hoi
display in red "Erro encontrado ao executar o ioh com o outcome `o', uf = `c', ano = `y'"
exit c(rc) // SHOW ERROR CODE AND STOP
}
if c(rc) ==0 post handle (`c') (`y') (`r(hoi_1)') (`r(d_1)') (`r(p_1)')
}
}
}


postclose handle

It happens, however, I want to make a leverage analysis to detect influential observations but do not know how to do, since I have to deal with the set of observations for each combination state-year as a database (ie, are 270 databases), and besides, I do not squeegee logistic regressions for each state and year, but a complete code that is only completed when all 270 indexes (HOI) are calculated.
I would like the help of Statalist members to know how can I leverage analysis to detect influential observations.

Thanks in advance

Girlan