Stata Guide¶

Basic Usage¶

dlm depvar, exposure(varname) unit(varname) time(varname) from(#) to(#)

The dlm command estimates a distributed lag model on depvar, using exposure as the treatment indicator. Unit and time fixed effects are absorbed via reghdfe, and standard errors are clustered at the unit level.

Options¶

Option	Required	Description
`exposure(varname)`	Yes	Treatment/exposure variable (typically 0/1)
`unit(varname)`	Yes	Panel unit identifier
`time(varname)`	Yes	Time period variable (numeric, evenly spaced)
`from(#)`	Yes	Earliest relative period (negative integer)
`to(#)`	Yes	Latest relative period (positive integer)
`ref(#)`	No	Reference period (default: −1)
`covariates(varlist)`	No	Additional control variables
`addl_fes(varlist)`	No	Additional fixed effects beyond unit and time
`verbose`	No	Display progress information

Examples¶

Basic estimation¶

dlm_gen_data, n_groups(500) n_times(20) treat_prob(0.4) seed(42) clear

dlm outcome, exposure(post) unit(unit) time(time) from(-3) to(3)

Output:

------------------------------------------------------------------------
Distributed Lag Model (Schmidheiny & Siegloch 2023)
------------------------------------------------------------------------
  Outcome:    outcome
  Exposure:   post
  Window:     [-3, 3], ref = -1
  N obs:      7500
  N clusters: 500
------------------------------------------------------------------------

        Time        Coef          SE     95% CI lo     95% CI hi
------------------------------------------------------------------------
          -3   -0.063959    0.483545     -1.011708      0.883790
          -2    0.094669    0.480424     -0.846962      1.036299
          -1       (ref)           .             .             .
           0   -2.763973    0.461907     -3.669310     -1.858636
           1   -3.094282    0.520414     -4.114294     -2.074271
           2   -2.707691    0.554940     -3.795373     -1.620010
           3   -3.256921    0.426200     -4.092272     -2.421569
------------------------------------------------------------------------

Wider event window¶

dlm outcome, exposure(post) unit(unit) time(time) from(-5) to(5)

Custom reference period¶

dlm outcome, exposure(post) unit(unit) time(time) from(-3) to(3) ref(-2)

With covariates¶

If your dataset includes control variables, pass them with covariates():

* Generate data and add a covariate
dlm_gen_data, n_groups(500) n_times(20) treat_prob(0.4) seed(42) clear
gen double x1 = rnormal()

dlm outcome, exposure(post) unit(unit) time(time) from(-3) to(3) covariates(x1)

Additional fixed effects¶

To absorb fixed effects beyond the default unit and time FEs (e.g., region-by-year):

dlm outcome, exposure(post) unit(unit) time(time) from(-3) to(3) addl_fes(region)

Subsetting with if/in¶

dlm outcome if unit <= 250, exposure(post) unit(unit) time(time) from(-3) to(3)

Multiple outcomes¶

Stata convention is one outcome per call. Loop for multiple:

foreach var of varlist outcome1 outcome2 outcome3 {
    dlm `var', exposure(post) unit(unit) time(time) from(-3) to(3)
    matrix betas_`var' = e(betas)
}

Working with Results¶

Extract the betas matrix¶

dlm outcome, exposure(post) unit(unit) time(time) from(-3) to(3)
matrix b = e(betas)

* Access individual cells: b[row, col]
* Columns: 1=time_to_event, 2=coef, 3=se, 4=ci_lo, 5=ci_hi
display "Effect at t=0: " b[4, 2]  // 4th row = period 0 (for from=-3)

Export to CSV¶

dlm outcome, exposure(post) unit(unit) time(time) from(-3) to(3)
matrix b = e(betas)

* Convert to dataset and export
preserve
clear
local nrows = rowsof(b)
set obs `nrows'
gen time_to_event = .
gen coef = .
gen se = .
gen ci_lo = .
gen ci_hi = .
forvalues i = 1/`nrows' {
    replace time_to_event = b[`i', 1] in `i'
    replace coef = b[`i', 2] in `i'
    replace se = b[`i', 3] in `i'
    replace ci_lo = b[`i', 4] in `i'
    replace ci_hi = b[`i', 5] in `i'
}
export delimited using "dlm_results.csv", replace
restore

Plot results¶

dlm outcome, exposure(post) unit(unit) time(time) from(-3) to(3)
matrix b = e(betas)

* Convert betas to variables for plotting
preserve
clear
local nrows = rowsof(b)
set obs `nrows'
gen time_to_event = .
gen coef = .
gen ci_lo = .
gen ci_hi = .
forvalues i = 1/`nrows' {
    replace time_to_event = b[`i', 1] in `i'
    replace coef = b[`i', 2] in `i'
    replace ci_lo = b[`i', 4] in `i'
    replace ci_hi = b[`i', 5] in `i'
}

twoway (rcap ci_lo ci_hi time_to_event, lcolor(navy)) ///
       (scatter coef time_to_event, mcolor(navy) msymbol(circle)), ///
       yline(0, lpattern(dash) lcolor(gray)) ///
       xline(-0.5, lpattern(dash) lcolor(gray)) ///
       xtitle("Time to Treatment") ytitle("Coefficient") ///
       title("Event Study (DLM)") legend(off)
restore

Stata event-study plot

Data Preservation¶

dlm uses preserve/restore internally. Your original dataset is never modified — no temporary variables leak into your data, and the observation count remains unchanged.

Test Data Generator¶

dlm_gen_data creates balanced panel data for testing:

dlm_gen_data, n_groups(500) n_times(20) treat_prob(0.4) seed(42) clear

Option	Default	Description
`n_groups(#)`	676	Number of panel units
`n_times(#)`	20	Number of time periods
`treat_prob(#)`	0.4	Fraction of units assigned to treatment
`seed(#)`	12345	Random seed for reproducibility
`clear`	—	Clear existing data in memory