Upgrading from csdid Version 1.82

Throughout this page, Version 1.82 means the last revision of the 1.8x line (November 2025), the version this page and the speed comparison were measured against. ssc install csdid installs Version 1.81, dated 2025-10-05; the two places where it differs from Version 1.82 are listed below.

The command surface is deliberately unchanged, so most do-files run exactly as they are. Four things can need your attention: results that move, postestimation forms that are gone, commands that are deprecated, and option spellings that have been renamed.

Results that move

Five defaults changed. The comparison group and the base period can change point estimates; so can trimming, where it binds, and balancing, on an unbalanced panel (see Unbalanced panels below). Inference changes standard errors and intervals, never point estimates.

	Version 1.82	2.0.0	To keep the old behavior
comparison group	never-treated	not-yet-treated	`nevertreated`
base period	varying	universal	`base_period(varying)`
inference	analytical, pointwise	multiplier bootstrap, simultaneous bands	`analytical`
propensity-score trimming	none	comparison observations at .995 or above, without a message	`pscoretrim(1)`
unbalanced panels	each 2×2 balanced separately	balanced once, `bal(full)`	`bal(pair)`

The first two rows are also the two defaults on which 2.0.0 departs from R did 2.5.1, which shares Version 1.82’s comparison group and base period; the other three follow R. Set both options explicitly when comparing implementations, alongside the same sample, estimator, weights, and inference settings.

Inference leaves the estimand alone: standard errors are now the multiplier bootstrap with simultaneous confidence bands by default, where Version 1.82 reported pointwise analytical standard errors. analytical restores analytical standard errors; add pointwise to request fully analytical pointwise intervals for aggregations too. Otherwise their simultaneous critical value is still bootstrapped.

To reproduce a Version 1.82 run, state the options explicitly (none of them is implied by the others). Let’s load the county mortality panel used throughout this site:

import delimited using ///
    "https://raw.githubusercontent.com/pedrohcgs/JEL-DiD/50f4f18/data/county_mortality_data.csv", ///
    clear varnames(1) bindquote(strict) stringcols(_all)
destring deaths population_20_64 year yaca county_code stfips, replace force
generate double mrate = 100000 * deaths / population_20_64
drop if missing(mrate) | population_20_64 <= 0
generate int gvar = yaca
replace gvar = 0 if missing(gvar) | gvar > 2019
bysort county_code: generate byte nyears = _N
keep if nyears == 11
save "jel_upgrade.dta", replace

use "jel_upgrade.dta", clear
csdid mrate, ivar(county_code) time(year) gvar(gvar) ///
    nevertreated base_period(varying) analytical pscoretrim(1)
display "comparison group: " e(control_group)
display "base period:   " e(base_period)
estat event

Those options, with bal(pair) added on an unbalanced panel, give you Version 1.82’s estimates and standard errors, with the exceptions listed in the next section. Drop them and you get the 2.0.0 defaults instead:

use "jel_upgrade.dta", clear
csdid mrate, ivar(county_code) time(year) gvar(gvar) rseed(20250101)
display "comparison group: " e(control_group)
display "base period:   " e(base_period)
estat event

Changes that options do not undo

These move numbers in a Version 1.82 do-file, and no option brings the old number back.

notyet pre-treatment cells follow Version 1.82’s asinr rule. In a pre-treatment cell with a varying base period, Version 1.82’s notyet compared cohort g only with cohorts treated after g; asinr compared it with every cohort not yet treated at t. Version 2.0.0’s notyet keeps a comparison unit when it is untreated in both periods of the comparison, which is the asinr rule, so under notyet base_period(varying) the pre-treatment cells, pre-period event-study coefficients and the pre-test differ from Version 1.82’s plain notyet. Post-treatment cells do not.
long reports pre-treatment cells with the opposite sign. Version 1.82’s long and long2 gave the same pre-treatment cells with opposite signs, and 2.0.0 gives long2’s for both. After long, every pre-treatment ATT(g,t) changes sign, and each pre-period event-study coefficient changes sign and moves one event time earlier: Version 1.82’s Tm1 is 2.0.0’s Tm2 with the sign reversed, and 2.0.0’s Tm1 is the reference row, fixed at 0.
The bootstrap draws Rademacher multipliers. Version 1.82’s wboot drew Mammen multipliers unless wbtype(rademacher) was given; 2.0.0 draws only Rademacher multipliers, so bootstrap standard errors and bands change as any change of draws changes them.
The overall standard error of estat group. The cohort effects, their standard errors and the overall effect are Version 1.82’s; the overall effect’s standard error is not (0.0119 against 0.0118 on the package’s mpdta example with lpop as covariate, and a wider gap under sampling weights). The event-study, calendar and simple aggregations report Version 1.82’s standard errors.
Time-varying iweights move the aggregations on a balanced panel. The cohort shares behind the simple, calendar and event-study aggregations, and behind the overall effect of estat group, count each unit at its iweight in the first period; Version 1.82 counted it at its mean iweight over its periods. Where a unit’s iweight varies over time, those estimates and their standard errors differ from Version 1.82’s, including the ones the previous item lists as unchanged. The two rules agree when each unit’s iweight is constant, and on an unbalanced panel under bal(pair), which also uses the mean. The ATT(g,t), their standard errors and the per-cohort effects of estat group do not move.

One further change moves no number but can stop a nevertreated run that Version 1.82 estimated. Version 1.82 had no size check on the never-treated group; 2.0.0 refuses a never-treated comparison group smaller than #covariates + 5 (measured as rows divided by periods), on balanced panels too. notyet, the default, does not depend on the never-treated group.

Version 1.81

ssc install csdid installs Version 1.81, which differs from Version 1.82 in two places. Its method(ipw) weighted without normalizing the weights; Version 1.82 and 2.0.0 normalize them (Version 1.81’s method(stdipw)), so method(ipw) estimates change and 2.0.0 has no unnormalized version. And aggregation standard errors on repeated cross sections differ slightly from Version 1.81’s. Everything else on this page applies to Version 1.81 as well.

Unbalanced panels

Version 1.82 balanced each 2×2 comparison separately, dropping the units missing from either of its periods: it said the panel was unbalanced, but not which or how many units each comparison dropped. csdid 2.0.0 makes the choice explicit with bal() and reports whatever it drops (units and observations both):


`bal(full)`	drop units not observed in every period, once, for all comparisons. Default, matching R `did`.
`bal(none)`	keep every unit and use the repeated-cross-section computation.
`bal(pair)`	Version 1.82’s per-comparison balancing: each 2x2 keeps the units observed in both of its periods.

bal(pair) reproduces Version 1.82’s estimand exactly, so an unbalanced-panel result from that version can be reproduced here by asking for it.

Postestimation forms that are gone

Each of these ran in Version 1.82 and stops a do-file in 2.0.0.

Version 1.82	2.0.0
`estat pretrend`	`csdid` prints the pre-test below the ATT(g,t) table and stores it in `e(wald_stat)`, `e(wald_df)` and `e(wald_pvalue)`. `window()` has no equivalent
`estat cevent`	No equivalent. `estat event, window(# #)` reports the effects inside an event-time window
`estat all`	Run `estat simple`, `estat group`, `estat calendar` and `estat event`
`estore()`, `esave()`	`estat` type`, post`, then `estimates store` or `estimates save`
`estat event, balance(#)`	`csdid_stats event, balance(#)`
`csdid ..., agg(simple)`, `agg(group)`, `agg(calendar)` or `agg(attgt)`	`csdid`, then `estat` type. `agg(event)` still works
`csdid, version`	`csdid version`
`csdid_stats attgt`, `csdid_stats cevent`	`estat attgt`; `cevent` has no equivalent
`csdid_stats` with no type	Aggregates by group, where Version 1.82 redisplayed the active results. Name the type
`csdid_stats ..., wboot`	Bootstrap inference is chosen when `csdid` runs; aggregation from a saved RIF file is analytical
`csdid_stats ..., save`	`storeall` keeps the aggregation’s influence functions in `e(agg_inffunc)`
`csdid_stats ..., post`	`estat` type`, post`
A RIF file written by Version 1.82	`csdid_stats using` reads files written by 2.0.0’s `saverif()`; run `csdid` again with `saverif()`. `csdid_stats` aggregates a file only through `using`, not after `use`

Coefficient names and stored results have a new layout, documented in help csdid: ATT(g,t) cells are named like g2004___2005_2003 (Version 1.82: g2004:t_2003_2005); the overall effect of estat group and estat calendar is Overall (Version 1.82: GAverage, CAverage); estat event reports no Pre_avg; and e(b), e(V), e(attgt) and e(N_clusters) take the place of Version 1.82’s e(b_attgt), e(V_attgt), e(gtt) and e(N_clust).

Deprecated commands

These still ship and still run, and each of them now prints a notice (a notice, not an error) saying what to use instead.

Deprecated	Use instead
`csdid_rif`	`estat attgt, saving(results) replace` then `use results, clear`
`csdid_table`	the table `csdid` prints, or `estat tidy, saving()` for the numbers
`dipt`	never documented; no replacement
`tsvmat`	never documented; no replacement

csgvar is not deprecated, since it builds the gvar() cohort variable from a treatment indicator and is fully supported.

The saved-RIF workflow itself is also still supported: csdid_stats using filename aggregates a file written by saverif(), and only the table-building command around it is deprecated. csdid_rif reads RIF files written by Version 1.82 and refuses those written by 2.0.0.

help csdid_legacy documents all of this inside Stata (offline), so you don’t have to keep this page open beside you.

Renamed options

Every Version 1.82 spelling below is accepted and says what to use instead, so you find out by running your do-file. Nothing on the list will stop a do-file from running, but long changes the sign of its pre-treatment cells (see above).

Old (warns)	Current
`long`, `long2`	`base_period(universal)`, which is now the default. Both give Version 1.82’s `long2` cells, so `long` reverses the sign of its Version 1.82 pre-treatment cells
`method(dripw)`	`method(dr)`
`method(stdipw)`	`method(ipw)`
`asinr`	no-op; use `notyet`
`never`	`nevertreated`. Not a no-op: the not-yet-treated comparison group is the default now, so this changes which units the treated are compared against

Spellings that are not renames

These are alternative names for current options, so they are not deprecated, nothing warns, and there is nothing at all to change in a do-file that uses them.

Spelling	Same as
`baseperiod()`	`base_period()`
`id()`	`ivar()`
`vce(cluster var)`	`cluster(var)`
`fixweights(base)`	`fix_weights(base_period)`
`balance()`	`bal()`, the same option unabbreviated
`unbalanced`	`bal(none)`. Typed in full – `unbal` is not an option
`allowunbalanced`, `allow_unbalanced`	`bal(none)` as well; the R-style longhand, also typed in full

Spellings that were never options

You may have seen these somewhere, but they have never been options in any release. They are not in Version 1.82, and 2.0.0 is the first release of the rewrite. csdid refuses them the way it refuses any other name it does not know.

Not an option	Type instead
`bal(unbal)`, `bal(unbalanced)`, `bal(allow_unbalanced)`	`bal(none)`, or the `unbalanced` option
`balanceall`, `bal(all)`	`bal(full)`
`balancepair`	`bal(pair)`
`lean`, `performance(...)`	nothing: influence functions stay internal at every size, and `storeall` is the one switch that copies them into `e()`

Options that now refuse

Each of these ran under Version 1.82 and stops a do-file in 2.0.0. We would rather stop a run than return a number we cannot vouch for, or one from a method this version does not offer.

Option	What happens now
`method(drimp)`	Errors. The improved doubly robust estimator is not offered; `method(dr)` is Version 1.82’s `method(dripw)`
`wboot(wbtype(mammen))`	Errors. Only the Rademacher multiplier is supported
`wboot(reps(#))` with `#` ≤ 20	Errors. Too few iterations to support a simultaneous band
`pscoretrim(#)` with `#` ≤ 0	Errors. Omit for the default `.995`, or pass `1` for no trimming
`gvar()` with negative values	Errors. Use `0` for never-treated
`from()`	Removed. It set a lower event-time bound on the simple, group and calendar aggregations, and R fixes that bound at event time 0, so there is no equivalent. `from(0)` was the legacy default and is already what `csdid` does, so most uses were no-ops. For event-time windows use `estat event, window(# #)`
`dryrun`	Rejected; it was an internal option

What you gain

There is no external dependency (no drdid, and nothing else). The SSC entry for Version 1.81 reads Requires: Stata version 14 and drdid from SSC, and Version 1.82 needs drdid too, while 2.0.0 needs nothing beyond Stata itself. It is also faster: between 10x and 35x, and never slower, on the fixed-size workloads in the package README, and from 10x to 308x as the design’s size, periods, cohorts and sampling scheme vary, on their own page.

Back to the guides.