<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>cdid — Difference-in-differences with incomplete panels | David Benatia</title><link>https://www.davidbenatia.com/tools/cdid/</link><atom:link href="https://www.davidbenatia.com/tools/cdid/index.xml" rel="self" type="application/rss+xml"/><description>cdid — Difference-in-differences with incomplete panels</description><generator>Wowchemy (https://wowchemy.com)</generator><language>en-us</language><lastBuildDate>Mon, 16 Nov 2026 00:00:00 +0000</lastBuildDate><image><url>https://www.davidbenatia.com/media/icon_huf51d053dcee75059daeadc3a21b01f3b_32112_512x512_fill_lanczos_center_3.png</url><title>cdid — Difference-in-differences with incomplete panels</title><link>https://www.davidbenatia.com/tools/cdid/</link></image><item><title>cdid models and comparisons</title><link>https://www.davidbenatia.com/tools/cdid/capabilities/</link><pubDate>Mon, 16 Nov 2026 00:00:00 +0000</pubDate><guid>https://www.davidbenatia.com/tools/cdid/capabilities/</guid><description>&lt;p>&lt;code>cdid&lt;/code> combines familiar treatment-effect specifications with chained comparisons and explicit handling of selective observation. The scope of each match matters. A package can implement several estimators, and a match for one specification does not imply that every option in that package has been reproduced.&lt;/p>
&lt;p>This table describes the R implementation, version 1.3.0.0. The &lt;a href="https://www.davidbenatia.com/tools/cdid/stata/">Stata package 1.0.0.0&lt;/a> is available too. Its &lt;a href="https://www.davidbenatia.com/tools/cdid/stata/from-r-cdid/">correspondence guide&lt;/a> records option mappings and differences.&lt;/p>
&lt;p>Read the &lt;a href="https://www.davidbenatia.com/tools/cdid/guide/from-other-packages/">matched commands and results&lt;/a> alongside this summary. For observation assumptions, see &lt;a href="https://www.davidbenatia.com/tools/cdid/missingness/">what handling an incomplete panel requires&lt;/a>.&lt;/p>
&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/capabilities/index.json">Download JSON&lt;/a> · &lt;a href="https://www.davidbenatia.com/tools/cdid/capabilities/index.md">Download Markdown&lt;/a>&lt;/p>
&lt;p>R cdid 1.3.0.0 · Reviewed 2026-11-16 · &lt;a href="https://github.com/dvbn/cdid/tree/b595cffabf9739d04d6b335604195450894ce958">Versioned source&lt;/a>&lt;/p>
&lt;h2 id="capabilities">Available capabilities&lt;/h2>
&lt;h3 id="chaining">Connect overlapping outcome histories&lt;/h3>
&lt;p>Combine supported within-unit changes across date pairs by GMM, with their covariance. Identification requires valid observation and parallel-trends restrictions, suitable controls, and sufficient support.&lt;/p>&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/examples/incomplete-panels/">Read the guide&lt;/a>&lt;/p>
&lt;h3 id="observation-adjustment">Adjust for selective observation&lt;/h3>
&lt;p>Choose observation weighting, actual-change imputation, or augmented reconstruction under conditional representativeness of mean changes and the documented model, support, and clipping conditions.&lt;/p>&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/missing-outcomes/">Read the guide&lt;/a>&lt;/p>
&lt;h3 id="treatment-adjustment">Choose treatment IPW, OR, or DR separately&lt;/h3>
&lt;p>Treatment weighting, counterfactual outcome regression, and doubly robust treatment adjustment are separate from the missing-outcome adjustment. The built-in conditional observation routes support all nine combinations.&lt;/p>&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/treatment/">Read the guide&lt;/a>&lt;/p>
&lt;h3 id="study-population">Declare the study population&lt;/h3>
&lt;p>For cohort/adoption fits, an external unit table can include units observed once or never. Required cohort, covariate, weight and cluster information depends on the selected adjustment. The static switching route has separate data requirements.&lt;/p>&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/help/prepare-data/">Read the guide&lt;/a>&lt;/p>
&lt;h3 id="cross-pair">Chain across different cohort observation windows&lt;/h3>
&lt;p>Cross-pair comparisons can connect target and control changes measured over different intervals under the declared control parallel-trends restrictions. chain_report() identifies contributing comparisons and unavailable cells.&lt;/p>&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/cross-pair-chaining/">Read the guide&lt;/a>&lt;/p>
&lt;h3 id="static-switching">Estimate static binary switching effects&lt;/h3>
&lt;p>Use current binary treatment status for effects at entry and exit under no carryover. This route has its own input requirements and switcher-weighted aggregate; it is not a dynamic switching model.&lt;/p>&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/static-entry-exit/">Read the guide&lt;/a>&lt;/p>
&lt;h2 id="assumption-connections">How the missingness assumptions relate&lt;/h2>
&lt;h3>The foundational chained-DiD framework&lt;/h3>&lt;p>Bellégo, Benatia and Dortet-Bernadet’s chained DiD paper provides the foundational framework for chained difference-in-differences with incomplete outcome histories. It explicitly formalizes missing-trends assumptions, distinguishes representativeness of outcome changes from representativeness of levels, and establishes identification, estimation and inference using overlapping incomplete panels. Its conditional extensions address selection on observed characteristics. This framework underlies cdid’s treatment of missing outcomes.&lt;/p>&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/missingness/#the-foundational-chained-did-framework">Explanation and sources&lt;/a>&lt;/p>&lt;h3>Linear fixed effects and mean-change representativeness&lt;/h3>&lt;p>In an additive linear model with correctly specified conditional mean changes, residuals with mean zero at every date conditional jointly on cohort, covariates, the unit effect, and the full observation history imply conditional mean representativeness of observed pairwise changes. This supplies a model-based sufficient condition for the mean form of the chained-DiD missing-trends restriction. Pairwise mean restrictions and full-history exogeneity are not identical, and selection on covariates requires the appropriate population averaging. This is a sufficient reconstruction of the connection, not a formal observation-history assumption explicitly stated in Wooldridge (2025), section 10.2.&lt;/p>&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/missingness/#4-the-connection-to-mtcar-in-linear-fixed-effects-models">Explanation and sources&lt;/a>&lt;/p>
&lt;h2 id="comparison-table">Which estimators can be reproduced?&lt;/h2>
&lt;div class="cdid-table">&lt;table>&lt;thead>&lt;tr>&lt;th>Method or software&lt;/th>&lt;th>Relationship&lt;/th>&lt;th>Scope of the match&lt;/th>&lt;/tr>&lt;/thead>&lt;tbody>
&lt;tr>&lt;td>&lt;a href="#did">Callaway-Sant&amp;#39;Anna: R did&lt;/a>&lt;/td>&lt;td>Direct delegation&lt;/td>&lt;td>mode = &amp;#39;ordinary&amp;#39; delegates to did, including repeated cross sections. Chained mode agrees only in the documented matched cases, after aligning sample, comparison group, covariates, reference period, and aggregation.&lt;/td>&lt;/tr>&lt;tr>&lt;td>&lt;a href="#csdid">Callaway-Sant&amp;#39;Anna: Stata csdid&lt;/a>&lt;/td>&lt;td>Matched implementation comparison&lt;/td>&lt;td>Use the corresponding R calculation in ordinary mode or a documented matched chained case. Align the two-date samples, controls, nuisance fits, reference periods, and aggregation; cdid does not delegate to Stata.&lt;/td>&lt;/tr>&lt;tr>&lt;td>&lt;a href="#linear-etwfe-jwdid">Wooldridge: linear etwfe and jwdid&lt;/a>&lt;/td>&lt;td>Matched linear equivalence&lt;/td>&lt;td>Use est_method = &amp;#39;reg&amp;#39; with the cohort_time_fe preset. Match unit and date effects, unrestricted treated cohort-date effects, never-treated or not-yet-treated controls, and optional time-invariant covariates. Requires a never-treated cohort, equal unit weights, and zero anticipation.&lt;/td>&lt;/tr>&lt;tr>&lt;td>&lt;a href="#sun-abraham">Sun-Abraham: sunab and eventstudyinteract&lt;/a>&lt;/td>&lt;td>Matched specification and recorded benchmark&lt;/td>&lt;td>The cohort_time_fe preset with never-treated controls recovers the saturated linear cohort-event specification with the same reference period. The guide records complete-panel post-treatment cell comparisons against R fixest::sunab.&lt;/td>&lt;/tr>&lt;tr>&lt;td>&lt;a href="#gardner-bjs">Gardner and Borusyak-Jaravel-Spiess&lt;/a>&lt;/td>&lt;td>Complete-panel point-estimate equivalence&lt;/td>&lt;td>On a complete unweighted panel, use the cohort_time_fe preset, not-yet-treated controls, PT-all from the first date, and unrestricted post-treatment cohort-date effects. Match the untreated unit-and-date regression, optionally interacting baseline covariates with dates, and aggregate with the same weights.&lt;/td>&lt;/tr>&lt;tr>&lt;td>&lt;a href="#lp-did">Dube-Girardi-Jorda-Taylor: LP-DiD&lt;/a>&lt;/td>&lt;td>Restricted long-difference equivalence&lt;/td>&lt;td>Match the unweighted clean-control long difference for a chosen baseline and horizon in the documented complete-panel, single-adoption-cohort design with never-treated controls. A pooled staggered comparison additionally requires identical risk sets and horizon weights.&lt;/td>&lt;/tr>&lt;tr>&lt;td>&lt;a href="#harmon-stepwise">Harmon: stepwise DiD&lt;/a>&lt;/td>&lt;td>Consecutive-link construction&lt;/td>&lt;td>Estimate and sum consecutive two-date DiDs for one target cohort, using the cohorts still untreated at each step&amp;#39;s endpoint. A standard chained fit can also use longer pairs and other supported control pools, so it need not give the same estimate.&lt;/td>&lt;/tr>&lt;tr>&lt;td>&lt;a href="#static-switchers">DIDmultiplegtSTAT and did_multiplegt_stat&lt;/a>&lt;/td>&lt;td>Matched binary static benchmark&lt;/td>&lt;td>Supply current binary treatment through dname and compare the documented switcher summaries on complete panels without covariates. The static model requires no carryover; this comparison does not cover dynamic treatment histories or the upstream packages&amp;#39; full treatment scope.&lt;/td>&lt;/tr>&lt;tr>&lt;td>&lt;a href="#cohort-chained">Balla-Elliott-Norwich: Cohort-Chained DiD&lt;/a>&lt;/td>&lt;td>Matched cohort-regression equivalence&lt;/td>&lt;td>Use not-yet-treated controls, universal base period, cross_pair = TRUE, pt_assumption = &amp;#39;ctrl&amp;#39;, and gmm_weighting = &amp;#39;within&amp;#39;. Recover the stacked cohort-fixed-effects point estimates when observation dates are common within each cohort and the regression sample and weights agree.&lt;/td>&lt;/tr>
&lt;/tbody>&lt;/table>&lt;/div>
&lt;h3 id="did">Callaway-Sant&amp;#39;Anna: R did&lt;/h3>
&lt;p>did balances the full panel by default; allowing an unbalanced panel uses repeated-cross-section computations. Chained mode instead combines supported within-unit changes and offers separate observation adjustments.&lt;/p>&lt;p>&lt;strong>Inference.&lt;/strong> Ordinary mode delegates inference to did. A chained comparison additionally requires matching the covariance and aggregation conventions; matching point estimates alone does not establish matching inference.&lt;/p>
&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/from-did/">Commands, assumptions and evidence&lt;/a> · &lt;a href="https://bcallaway11.github.io/did/reference/att_gt.html">Source&lt;/a> · &lt;a href="https://doi.org/10.1016/j.jeconom.2020.12.001">Source&lt;/a>&lt;/p>
&lt;h3 id="csdid">Callaway-Sant&amp;#39;Anna: Stata csdid&lt;/h3>
&lt;p>csdid 2.0.3 defaults to bal(full); bal(pair) retains units observed at both comparison dates, and bal(none) uses repeated-cross-section computations. Pair balancing is the legacy 1.82 observation rule. Chaining can use additional supported pairs.&lt;/p>&lt;p>&lt;strong>Inference.&lt;/strong> Match variance, clustering, bootstrap, confidence-band, and aggregation conventions separately. No universal R-Stata inference equivalence is asserted.&lt;/p>
&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/from-other-packages/">Commands, assumptions and evidence&lt;/a> · &lt;a href="https://psantanna.com/csdid/articles/unbalanced-panels.html">Source&lt;/a> · &lt;a href="https://psantanna.com/csdid/articles/comparing-did-commands.html">Source&lt;/a>&lt;/p>
&lt;h3 id="linear-etwfe-jwdid">Wooldridge: linear etwfe and jwdid&lt;/h3>
&lt;p>The documented match covers complete and incomplete panels on the same estimation rows. Use unit effects when observation dates vary within cohorts. Explicit observation correction is an additional cdid specification, outside the uncorrected regression match.&lt;/p>&lt;p>&lt;strong>Inference.&lt;/strong> Matched cell effects and covariance require common effect-evaluation covariates, clustering, and sandwich normalization. Covariate-mean uncertainty and event-study aggregation weights must also be aligned; critical values are a separate choice.&lt;/p>
&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/cohort-time-fe/#match-r-etwfe-and-stata-jwdid">Commands, assumptions and evidence&lt;/a> · &lt;a href="https://grantmcdermott.com/etwfe/reference/etwfe.html">Source&lt;/a> · &lt;a href="https://github.com/friosavila/stpackages/blob/main/jwdid/jwdid.sthlp">Source&lt;/a> · &lt;a href="https://doi.org/10.1007/s00181-025-02807-z">Source&lt;/a>&lt;/p>
&lt;h3 id="sun-abraham">Sun-Abraham: sunab and eventstudyinteract&lt;/h3>
&lt;p>The recorded benchmark is on complete panels. It does not establish a general incomplete-panel or selective-observation equivalence with either upstream command.&lt;/p>&lt;p>&lt;strong>Inference.&lt;/strong> The stated benchmark concerns post-treatment cell point estimates. Event-study averages require the same cohort weights; covariance and inference are separate comparisons.&lt;/p>
&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/from-other-packages/#matched-specifications-in-cdid">Commands, assumptions and evidence&lt;/a> · &lt;a href="https://lrberge.github.io/fixest/reference/sunab.html">Source&lt;/a> · &lt;a href="https://github.com/lsun20/EventStudyInteract">Source&lt;/a> · &lt;a href="https://doi.org/10.1016/j.jeconom.2020.09.006">Source&lt;/a>&lt;/p>
&lt;h3 id="gardner-bjs">Gardner and Borusyak-Jaravel-Spiess&lt;/h3>
&lt;p>The recorded matches use complete panels. With incomplete outcomes, joint fixed-effects fitting and untreated-only imputation can use different information to estimate the untreated path. Actual-change imputation for selective observation is a separate operation.&lt;/p>&lt;p>&lt;strong>Inference.&lt;/strong> The guide records post-treatment point-estimate matches to R did2s and didimputation. No general covariance or confidence-interval equivalence is claimed.&lt;/p>
&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/from-other-packages/#two-stage-and-imputation-event-studies">Commands, assumptions and evidence&lt;/a> · &lt;a href="https://kylebutts.github.io/did2s/">Source&lt;/a> · &lt;a href="https://github.com/kylebutts/didimputation">Source&lt;/a> · &lt;a href="https://github.com/borusyak/did_imputation">Source&lt;/a> · &lt;a href="https://doi.org/10.1093/restud/rdae007">Source&lt;/a>&lt;/p>
&lt;h3 id="lp-did">Dube-Girardi-Jorda-Taylor: LP-DiD&lt;/h3>
&lt;p>The illustrated equivalence does not establish an incomplete-panel match. Other LP-DiD weighting and baseline options are outside this comparison.&lt;/p>&lt;p>&lt;strong>Inference.&lt;/strong> The claim is a point-estimate match for the stated long difference. It does not certify the complete LP-DiD interface or all variance and inference options.&lt;/p>
&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/from-other-packages/#two-date-comparison-with-r-did">Commands, assumptions and evidence&lt;/a> · &lt;a href="https://github.com/danielegirardi/lpdid">Source&lt;/a> · &lt;a href="https://doi.org/10.1002/jae.70000">Source&lt;/a>&lt;/p>
&lt;h3 id="harmon-stepwise">Harmon: stepwise DiD&lt;/h3>
&lt;p>For an incomplete-panel comparison, match Harmon&amp;#39;s target units eligible for the requested horizon and the observed controls in every step. Merely summing arbitrary pair-specific estimates does not establish the match.&lt;/p>&lt;p>&lt;strong>Inference.&lt;/strong> The guide reproduces the point-estimate construction. Uncertainty for the sum must include covariance across steps because units can contribute to several comparisons.&lt;/p>
&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/from-other-packages/#harmons-stepwise-did">Commands, assumptions and evidence&lt;/a> · &lt;a href="https://web.econ.ku.dk/nharmon/docs/harmon2024onlineappendix.pdf">Source&lt;/a> · &lt;a href="https://web.econ.ku.dk/nharmon/docs/harmon2022difference.pdf">Source&lt;/a>&lt;/p>
&lt;h3 id="static-switchers">DIDmultiplegtSTAT and did_multiplegt_stat&lt;/h3>
&lt;p>The recorded upstream benchmark uses complete panels. cdid additionally supports chained static estimation with incomplete outcomes under its observation and identification restrictions.&lt;/p>&lt;p>&lt;strong>Inference.&lt;/strong> Use the guide&amp;#39;s exact switcher aggregation and reference commands. The benchmark does not establish universal covariance or inference equivalence across STAT options.&lt;/p>
&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/from-other-packages/#binary-static-switcher-summaries">Commands, assumptions and evidence&lt;/a> · &lt;a href="https://cran.r-project.org/package=DIDmultiplegtSTAT">Source&lt;/a> · &lt;a href="https://github.com/Credible-Answers/did_multiplegt_stat">Source&lt;/a> · &lt;a href="https://arxiv.org/abs/2201.06898v7">Source&lt;/a>&lt;/p>
&lt;h3 id="cohort-chained">Balla-Elliott-Norwich: Cohort-Chained DiD&lt;/h3>
&lt;p>The matched cohort regression has common observation dates within cohorts. cdid&amp;#39;s cross-pair implementation also supports within-individual changes with different observation dates and explicit selective-observation corrections under its stated assumptions.&lt;/p>&lt;p>&lt;strong>Inference.&lt;/strong> The stated equivalence is for point estimates. Match sampling units, clusters, aggregation, and covariance conventions separately before comparing uncertainty.&lt;/p>
&lt;p>&lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/cross-pair-chaining/#options-for-chaining">Commands, assumptions and evidence&lt;/a> · &lt;a href="https://pdfs.dballaelliott.com/ccdid.pdf">Source&lt;/a> · &lt;a href="https://doi.org/10.1016/j.jeconom.2024.105783">Source&lt;/a>&lt;/p></description></item><item><title>Incomplete panels: what chaining and observation correction add</title><link>https://www.davidbenatia.com/tools/cdid/missingness/</link><pubDate>Mon, 16 Nov 2026 00:00:00 +0000</pubDate><guid>https://www.davidbenatia.com/tools/cdid/missingness/</guid><description>&lt;p>&lt;strong>&lt;code>cdid&lt;/code> connects the outcome changes your panel supplies and makes explicit when those changes represent your study population.&lt;/strong> It also provides observation weighting, actual-change imputation, and augmented reconstruction when observation depends on measured characteristics. These choices work separately from the adjustment that makes treated units comparable with controls.&lt;/p>
&lt;h2 id="the-foundational-chained-did-framework">The foundational chained-DiD framework&lt;/h2>
&lt;p>Bellégo, Benatia and Dortet-Bernadet’s chained DiD paper provides the foundational framework for chained difference-in-differences with incomplete outcome histories. It explicitly formalizes missing-trends assumptions, distinguishes representativeness of outcome changes from representativeness of levels, and establishes identification, estimation and inference using overlapping incomplete panels. Its conditional extensions address selection on observed characteristics. This framework underlies cdid’s treatment of missing outcomes. &lt;a href="https://arxiv.org/html/2301.01085v4#S2.SS2" target="_blank" rel="noopener">The Chained Difference-in-Differences, section 2.2&lt;/a> · &lt;a href="https://doi.org/10.1016/j.jeconom.2024.105783" target="_blank" rel="noopener">Published paper&lt;/a>&lt;/p>
&lt;p>Consider households interviewed at dates 1 and 2 and other households interviewed at dates 2 and 3. Few households appear at both endpoints of the longer interval. Chaining uses the two shorter comparisons to recover a longer effect when the observation and parallel trends assumptions support them. Additional valid pairs can contribute too. The package combines their information by GMM, accounting for shared units and estimated models. &lt;a href="https://www.davidbenatia.com/tools/cdid/guide/examples/incomplete-panels/">Worked example&lt;/a>&lt;/p>
&lt;p>Several other DiD packages accept incomplete panels. The useful comparison concerns which observations enter each estimate, which population it describes, and which assumptions justify it. Six distinctions clarify what &lt;code>cdid&lt;/code> adds.&lt;/p>
&lt;h2 id="1-r-did-can-use-an-unbalanced-panel">1. R did can use an unbalanced panel&lt;/h2>
&lt;p>The default &lt;code>allow_unbalanced_panel = FALSE&lt;/code> drops units not observed in every period. With unbalanced panels allowed, R &lt;code>did&lt;/code> retains incomplete histories and uses repeated-cross-section computations. Its documentation distinguishes changes within a unit, conditional on earlier covariates, from period-specific outcome calculations with contemporaneous covariates. &lt;a href="https://bcallaway11.github.io/did/reference/att_gt.html" target="_blank" rel="noopener">R did reference&lt;/a>&lt;/p>
&lt;p>This is a legitimate estimation route under its assumptions. It does not construct the same information as chaining within-unit changes across overlapping histories. In &lt;code>cdid&lt;/code>, &lt;code>mode = &amp;quot;ordinary&amp;quot;&lt;/code> delegates to &lt;code>did&lt;/code>, including its repeated-cross-section route; chained mode performs the within-unit calculation. &lt;a href="https://www.davidbenatia.com/tools/cdid/guide/from-did/">Moving from did&lt;/a>&lt;/p>
&lt;h2 id="2-pair-balancing-supplies-a-change-not-an-observation-correction">2. Pair balancing supplies a change, not an observation correction&lt;/h2>
&lt;p>Stata &lt;code>csdid&lt;/code> 2.0.3 offers three choices. &lt;code>bal(full)&lt;/code>, the default, retains units observed throughout. &lt;code>bal(pair)&lt;/code> requires both outcomes separately for each two-date comparison. &lt;code>bal(none)&lt;/code> uses repeated-cross-section computations with available observations. Pair balancing reproduces the observation rule of legacy version 1.82. &lt;a href="https://psantanna.com/csdid/articles/unbalanced-panels.html" target="_blank" rel="noopener">Official unbalanced-panel guide&lt;/a>&lt;/p>
&lt;p>Having two observed outcomes makes their difference available. It does not establish that its mean represents the original cohort. A baseline-to-follow-up comparison can also be unavailable while shorter, overlapping changes remain informative. Chaining can connect those changes; an observation correction addresses their selectivity. These solve different problems. &lt;a href="https://www.davidbenatia.com/tools/cdid/guide/from-other-packages/">Matched comparisons&lt;/a>&lt;/p>
&lt;h2 id="3-linear-etwfe-and-jwdid-already-support-incomplete-panels">3. Linear etwfe and jwdid already support incomplete panels&lt;/h2>
&lt;p>Their regression specifications can use incomplete outcome histories. The distinction between unit and cohort fixed effects matters: their balanced-panel equivalence need not survive when observation dates differ within cohorts. R &lt;code>etwfe&lt;/code> uses cohort effects when &lt;code>ivar&lt;/code> is omitted; its unit-effects specification needs the identifier. Stata &lt;code>jwdid&lt;/code> documents unit effects and its &lt;code>group&lt;/code>, &lt;code>corr&lt;/code>, and &lt;code>cre&lt;/code> alternatives. &lt;a href="https://grantmcdermott.com/etwfe/reference/etwfe.html" target="_blank" rel="noopener">etwfe reference&lt;/a> · &lt;a href="https://github.com/friosavila/stpackages/blob/main/jwdid/jwdid.sthlp" target="_blank" rel="noopener">jwdid help&lt;/a>&lt;/p>
&lt;p>The &lt;code>cdid&lt;/code> cohort-by-date preset reproduces the documented linear unit-effects comparisons on matched complete and incomplete panels. That match requires aligned samples, covariates, controls, effect evaluation, and covariance normalization. Event-study averages additionally require the same aggregation weights. &lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/cohort-time-fe/">Commands and verified scope&lt;/a>&lt;/p>
&lt;h2 id="4-the-connection-to-mtcar-in-linear-fixed-effects-models">4. The connection to MTCAR in linear fixed-effects models&lt;/h2>
&lt;p>Wooldridge&amp;rsquo;s section 10.2 describes a related selection mechanism within linear fixed-effects models. It permits missingness related to additive individual heterogeneity and rules out systematic selection on idiosyncratic errors, using verbal restrictions and a regression construction that draws on Wooldridge (2019). That subsection does not separately state a missing-trends assumption or an identification theorem for incomplete panels. &lt;a href="https://link.springer.com/article/10.1007/s00181-025-02807-z#Sec30" target="_blank" rel="noopener">Wooldridge (2025), section 10.2&lt;/a>&lt;/p>
&lt;p>The argument below makes the connection to the chained-DiD framework explicit. Its full-history mean-exogeneity condition is a sufficient reconstruction, not an assumption displayed in that subsection. With suitable restrictions, the additive linear model implies mean-change representativeness. When covariates affect both trends and observation, the corresponding restriction is conditional on those covariates.&lt;/p>
&lt;p>To see the connection, write an additive model as&lt;/p>
&lt;pre>&lt;code class="language-text">Y(i,t) = c(i) + m(t, cohort, X(i)) + u(i,t).
&lt;/code>&lt;/pre>
&lt;p>Differencing removes &lt;code>c(i)&lt;/code>. If the residual has mean zero at every date conditional on cohort, covariates, the unit effect and the observation history, then&lt;/p>
&lt;pre>&lt;code class="language-text">E[Y(i,t) - Y(i,s) | cohort, X(i), both dates observed]
= E[Y(i,t) - Y(i,s) | cohort, X(i)].
&lt;/code>&lt;/pre>
&lt;p>This is a sufficient-condition argument connecting the two formulations. Chained DiD states the restriction directly on outcome changes; it need not impose that stronger full-history restriction on the regression errors. Without covariate selection, this gives baseline mean-change representativeness. With covariate selection, recovering a population effect also requires averaging over the population&amp;rsquo;s covariate distribution.&lt;/p>
&lt;p>The relevant changes include treatment gains. If observation selects heterogeneous gains not accounted for by the model, removing a unit effect does not resolve that selection. Thus the linear fixed-effects argument and the chained-DiD observation assumption share a mechanism, while their full assumptions and estimators need not coincide. The &lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/missing-outcomes/">missing-outcome guide&lt;/a> explains the additional adjustments available in &lt;code>cdid&lt;/code>.&lt;/p>
&lt;h2 id="5-chaining-requires-representative-mean-changes">5. Chaining requires representative mean changes&lt;/h2>
&lt;p>In the chained adoption model, with &lt;code>selection = &amp;quot;none&amp;quot;&lt;/code>, units observed at both dates must be representative of their cohort &lt;strong>in mean changes&lt;/strong> over each used interval:&lt;/p>
&lt;pre>&lt;code class="language-text">E[change | cohort, both dates observed] = E[change | cohort]
&lt;/code>&lt;/pre>
&lt;p>This concerns population expectations. Individual changes need not be identical, and sample means need not coincide exactly. Observation can select outcome levels while preserving the needed mean changes. The original paper explains why mean-independence suffices even though it states a stronger independence condition for exposition. &lt;a href="https://arxiv.org/html/2301.01085v4" target="_blank" rel="noopener">Chained DiD, discussion of Assumption 2&lt;/a>&lt;/p>
&lt;p>When representativeness holds only conditional on covariates, the adjustment must reconstruct population changes from those conditional comparisons. The restriction covers actual treated changes, including treatment effects, and every pair used. Connected observation windows alone do not establish identification. &lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/missing-outcomes/">Missing-outcome assumptions&lt;/a>&lt;/p>
&lt;p>The static switching model instead uses switcher and stayer populations. With covariates, its baseline also requires representativeness of their reference-covariate composition. See the &lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/static-entry-exit/">static-model assumptions&lt;/a>.&lt;/p>
&lt;h2 id="6-treatment-dr-does-not-automatically-correct-selective-observation">6. Treatment DR does not automatically correct selective observation&lt;/h2>
&lt;p>A treatment propensity and an untreated-outcome regression address the treatment comparison. An observation propensity and an actual-change model address which changes are observed. The latter model includes treated changes; it is different from predicting the untreated counterfactual.&lt;/p>
&lt;p>Under &lt;code>selection = &amp;quot;conditional_mtar&amp;quot;&lt;/code>, &lt;code>cdid&lt;/code> offers three adjustments:&lt;/p>
&lt;ul>
&lt;li>&lt;strong>Observation weighting:&lt;/strong> weight each observed change using its probability of being observed at both dates.&lt;/li>
&lt;li>&lt;strong>Change imputation:&lt;/strong> predict actual changes and average them over the declared study population.&lt;/li>
&lt;li>&lt;strong>Augmented correction:&lt;/strong> combine those predictions with weighted prediction errors from observed changes.&lt;/li>
&lt;/ul>
&lt;p>The paper&amp;rsquo;s Corollary 1 establishes the conditional observation-weighting argument. The package guide documents imputation and augmentation and their separate model requirements. Treatment IPW, OR, and DR remain distinct choices. &lt;a href="https://arxiv.org/html/2301.01085v4" target="_blank" rel="noopener">Corollary 1&lt;/a> · &lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/missing-outcomes/#which-models-must-be-correct">Adjustment guide&lt;/a>&lt;/p>
&lt;h2 id="define-the-population-before-choosing-the-fit">Define the population before choosing the fit&lt;/h2>
&lt;p>For the cohort/adoption route, supply &lt;code>frame&lt;/code> when the study includes units with no outcome rows. It retains their population information, with the cohort and covariates required by the selected adjustment. Probability-based corrections require observation support; imputation needs training and prediction support. Model restrictions, parallel trends, and no anticipation still matter. Binding probability safeguards can change the target, so the documented robustness claims retain their clipping qualifications. &lt;a href="https://www.davidbenatia.com/tools/cdid/guide/help/prepare-data/">Prepare the unit table&lt;/a> · &lt;a href="https://www.davidbenatia.com/tools/cdid/guide/methods/missing-outcomes/#probability-safeguards">Probability safeguards&lt;/a>&lt;/p>
&lt;p>&lt;strong>Comparison reviewed November 16, 2026.&lt;/strong> The R syntax and comparison evidence on this page describe &lt;code>cdid&lt;/code> 1.3.0.0. The &lt;a href="https://www.davidbenatia.com/tools/cdid/stata/">native Stata package&lt;/a> has its own guide and &lt;a href="https://www.davidbenatia.com/tools/cdid/stata/from-r-cdid/">R correspondence&lt;/a>. Upstream sources inspected were R &lt;code>did&lt;/code> master 2.5.1.906, Stata &lt;code>csdid&lt;/code> 2.0.3, R &lt;code>etwfe&lt;/code> online documentation with a changelog labeled 0.6.2.99 development, and Stata &lt;code>jwdid&lt;/code> source 2.201. The recorded &lt;code>cdid&lt;/code> regression comparisons used &lt;code>etwfe&lt;/code> 0.6.2; their exact settings remain in the linked guide. This source review does not rerun those numerical comparisons. &lt;a href="https://github.com/bcallaway11/did/blob/master/DESCRIPTION" target="_blank" rel="noopener">did version&lt;/a> · &lt;a href="https://psantanna.com/csdid/" target="_blank" rel="noopener">csdid release&lt;/a> · &lt;a href="https://grantmcdermott.com/etwfe/news/index.html" target="_blank" rel="noopener">etwfe changelog&lt;/a> · &lt;a href="https://github.com/friosavila/stpackages/blob/main/jwdid/jwdid.ado" target="_blank" rel="noopener">jwdid source&lt;/a>&lt;/p></description></item></channel></rss>