Question about correct implementation of command ivregress 2sls

[Michael Duarte Goncalves]

Hi everyone,

I really have a quick question about the command -ivregress 2sls-. I am bit confused about the right way of using it. Are those two loops identical, or not?

    eststo clear

    foreach var_pop in d_log_pop d_pop {
        foreach tech in wind solar {
            foreach region in ccaa nat {
                foreach fe in com_id prov_code {

                    local ss_var   = "ss_`tech'_`region'"
                    local d_cap    = "d_power_`tech'"
                    local fe_short = cond("`fe'" == "com_id", "cid", "pid")
                    local tag      = "`var_pop'_`tech'_`region'_`fe_short'"
                    
                    * Special-case installation capacity
                    if "`region'" == "nat" {
                        local cap_var = "new_inst_cap_`tech'_pen"
                    }
                    else if "`region'" == "ccaa" {
                        local cap_var = "new_inst_cap_`tech'_ccaa"
                    }

                    * Potential variable by tech
                    if "`tech'" == "solar" {
                        local potential_var_std = "com_pv_potential_std"
                    }
                    else if "`tech'" == "wind" {
                        local potential_var_std = "com_wind_power_dens_std"
                    }                        

                    * Run IV via ivreghdfe, absorbing FE + year
                    ivregress 2sls `var_pop' (`d_cap' = `ss_var') `cap_var' `potential_var_std' i.`fe' i.year, robust
                    eststo iv_`tag'

                    * Tag stored result with checkmarks for Year FE and respective FE
                    estadd local yearfe     "\ding{51}"
                    estadd local comarcafe  = cond("`fe'" == "com_id", "\ding{51}", "")
                    estadd local provincefe = cond("`fe'" == "prov_code", "\ding{51}", "")
                }
            }
        }
    }

    eststo clear

    foreach var_pop in d_log_pop d_pop {
        foreach tech in wind solar {
            foreach region in ccaa nat {
                foreach fe in com_id prov_code {

                    local ss_var   = "ss_`tech'_`region'"
                    local d_cap    = "d_power_`tech'"
                    local fe_short = cond("`fe'" == "com_id", "cid", "pid")
                    local tag      = "`var_pop'_`tech'_`region'_`fe_short'"
                    
                    * Special-case installation capacity
                    if "`region'" == "nat" {
                        local cap_var = "new_inst_cap_`tech'_pen"
                    }
                    else if "`region'" == "ccaa" {
                        local cap_var = "new_inst_cap_`tech'_ccaa"
                    }

                    * Potential variable by tech
                    if "`tech'" == "solar" {
                        local potential_var_std = "com_pv_potential_std"
                    }
                    else if "`tech'" == "wind" {
                        local potential_var_std = "com_wind_power_dens_std"
                    }                        

                    * Run IV via ivreghdfe, absorbing FE + year
                    ivregress 2sls `var_pop' (`d_cap' = `ss_var' `cap_var' `potential_var_std') i.`fe' i.year, robust
                    eststo iv_`tag'

                    * Tag stored result with checkmarks for Year FE and respective FE
                    estadd local yearfe     "\ding{51}"
                    estadd local comarcafe  = cond("`fe'" == "com_id", "\ding{51}", "")
                    estadd local provincefe = cond("`fe'" == "prov_code", "\ding{51}", "")
                }
            }
        }
    }

I put in bold green the main changes between loops.
Thank you very much for your help.

Edit: I noticed that results could be different. If I am not mistaken, the first loop tries to do something like:

regress `d_cap' `ss_var' `cap_var' `potential_var_std' i.`fe'  i.year, r

predict double d_cap_hat, xb

regress `var_pop' d_cap_hat `cap_var' `potential_var_std' i.`fe' i.year, robust

But I do not understand why. Thanks.

Best,
Michael

[George Ford]

You're treating more variables as exclusions in the second batch, so the results should be different.