{smcl} {* *! version 2.0.3 30oct2026}{...} {vieweralsosee "csdid" "help csdid"}{...} {vieweralsosee "csdid postestimation" "help csdid_postestimation"}{...} {vieweralsosee "csdid_stats" "help csdid_stats"}{...} {vieweralsosee "csdid_estat" "help csdid_estat"}{...} {vieweralsosee "" "--"}{...} {vieweralsosee "[G-2] graph twoway" "help twoway"}{...} {vieweralsosee "[G-2] graph twoway rcap" "help twoway_rcap"}{...} {vieweralsosee "[G-2] graph twoway rarea" "help twoway_rarea"}{...} {vieweralsosee "[G-3] by_option" "help by_option"}{...} {viewerjumpto "Syntax" "csdid_plot##syntax"}{...} {viewerjumpto "Description" "csdid_plot##description"}{...} {viewerjumpto "Options" "csdid_plot##options"}{...} {viewerjumpto "Remarks" "csdid_plot##remarks"}{...} {viewerjumpto "Plot-data schema" "csdid_plot##schema"}{...} {viewerjumpto "Notes on the exported series" "csdid_plot##plotnotes"}{...} {viewerjumpto "Relationship to legacy Stata csdid Version 1.82" "csdid_plot##legacy"}{...} {viewerjumpto "Examples" "csdid_plot##examples"}{...} {viewerjumpto "Stored results" "csdid_plot##results"}{...} {viewerjumpto "Diagnostics" "csdid_plot##errors"}{...} {viewerjumpto "References" "csdid_plot##references"}{...} {viewerjumpto "Authors" "csdid_plot##authors"}{...} {title:Title} {p2colset 5 20 22 2}{...} {p2col :{cmd:csdid_plot} {hline 2}}Draw or export plots of {cmd:csdid} group-time average treatment effects{p_end} {p2colreset}{...} {marker syntax}{...} {title:Syntax} {p 8 15 2} {cmd:csdid_plot} [{cmd:,} {opth sav:ing(filename)} {opt replace} {opt group(numlist)}] {pstd} Without {opt sav:ing()}, {cmd:csdid_plot} draws the graph; with {opt sav:ing()} it writes the plot data to a dataset and draws nothing. {pstd} {cmd:estat plot} is the same command under its {cmd:estat} spelling: the options are handed to {cmd:csdid_plot} unchanged, so the two forms are interchangeable after {cmd:csdid}. {pstd} The same syntax covers two plots. Run directly after {cmd:csdid}, it plots the ATT(g,t) estimates; run after an aggregation has been computed with {helpb csdid_stats} or {helpb csdid_estat}, it plots that aggregation. See {help csdid_plot##description:Description}. {synoptset 24 tabbed}{...} {synopthdr} {synoptline} {syntab:Main} {synopt :{opth sav:ing(filename)}}write the plot data to {it:filename} instead of drawing the graph{p_end} {synopt :{opt replace}}overwrite {it:filename} if it already exists{p_end} {syntab:Selection {help csdid_plot##opt_group:[+]}} {synopt :{opt group(numlist)}}restrict the plotted rows to the listed treatment cohorts; applies to ATT(g,t) and group-type aggregation plots only{p_end} {synoptline} {p2colreset}{...} {p 4 6 2} {cmd:csdid_plot} is a postestimation command and requires results from {helpb csdid} in {cmd:e()}.{p_end} {p 4 6 2} No other options are accepted; graph-styling options are rejected with an error. See {help csdid_plot##remarks:Remarks}.{p_end} {marker description}{...} {title:Description} {pstd} {cmd:csdid_plot} draws a Callaway and Sant'Anna (2021) event-study or group-time plot: the point estimates as markers, the confidence bands as capped spikes, a dashed line at zero, and the pre-treatment estimates in a different color from the post-treatment ones. The graph is drawn in the current scheme and takes no styling options. {pstd} {opt sav:ing()} writes the same picture as numbers instead. It exports a small Stata dataset containing everything the figure shows -- the point estimates, the confidence-band bounds, the x-axis values, and the pre-/post-treatment labels -- and draws nothing. That is the route to a figure of your own design: you draw it with {helpb twoway} (or any other tool), so the appearance is entirely under your control and the numbers behind it are inspectable, storable, and reproducible. {pstd} Which plot you get depends on what is currently in {cmd:e()}: {p2colset 5 40 42 2}{...} {p2col :{it:what is in e()}}{it:what is plotted}{p_end} {p2line} {p2col :{cmd:csdid} results, no aggregation}ATT(g,t) by cohort and calendar period{p_end} {p2col :{cmd:type(dynamic)} aggregation}event study: ATT by length of exposure{p_end} {p2col :{cmd:type(group)} aggregation}ATT by treatment cohort{p_end} {p2col :{cmd:type(calendar)} aggregation}ATT by calendar period{p_end} {p2col :{cmd:type(simple)} aggregation}nothing; {cmd:csdid_plot} exits with error 498{p_end} {p2line} {p2colreset}{...} {pstd} On the ATT(g,t) plot the cohorts are arranged in panels, one per treatment cohort, with the x axis running over calendar periods; on an aggregation plot the x axis is event time, cohort, or calendar period, as the case may be. {pstd} The rule is mechanical: if the matrix {cmd:e(aggte)} is present, the active aggregation is plotted; otherwise the ATT(g,t) table {cmd:e(attgt)} is plotted. {cmd:e(aggte)} is posted by {helpb csdid_stats}, by {cmd:estat dynamic}/{cmd:group}/{cmd:calendar}/{cmd:event} (see {helpb csdid_estat}), and by {cmd:csdid, agg(event)}. To go back to the ATT(g,t) plot after an aggregation, rerun {cmd:csdid}. {pstd} {cmd:csdid_plot} does not modify the data in memory (it uses {helpb preserve} and {helpb restore}) and does not modify {cmd:e()}. {marker options}{...} {title:Options} {dlgtab:Main} {phang} {opth sav:ing(filename)} switches {cmd:csdid_plot} from drawing to exporting. It names the Stata dataset that {cmd:csdid_plot} writes; {cmd:.dta} is added if {it:filename} has no extension. No graph is drawn when {opt sav:ing()} is specified. {pmore} Both spellings of {cmd:replace} work. {cmd:csdid_plot, saving(myplot) replace} is the documented form, and the more common Stata idiom {cmd:saving(myplot, replace)} is parsed as well: the comma inside {cmd:saving()} separates the filename from its sub-option and never becomes part of the filename. {cmd:replace} is the only sub-option {cmd:saving()} accepts; anything else after the comma is refused by name with return code 198, as is a comma with no filename in front of it. {phang} {opt replace} permits {cmd:csdid_plot} to overwrite an existing {it:filename}. Without it, writing over an existing file fails with the usual "file already exists" error, return code 602. On its own, without {cmd:saving()}, it means nothing and is refused with return code 198. {marker opt_group}{...} {dlgtab:Selection} {phang} {opt group(numlist)} restricts the plotted rows to the listed treatment cohorts, that is, to the listed values of the {cmd:gvar()} variable. It is useful when a design has many cohorts and you want one figure per subset. It applies to the drawn graph and to the {opt sav:ing()} export alike. A cohort must match exactly, and a {it:numlist} holds 13 significant digits, so a cohort that does not fit in 13 digits, such as a monthly date stored as 2000 + 7/12, cannot be selected; write the plot data with {opt sav:ing()} and select those rows there. {phang2} On the ATT(g,t) plot, {cmd:group()} keeps the rows whose {cmd:group} value is in {it:numlist}. {phang2} On a {cmd:type(group)} aggregation plot, {cmd:group()} keeps the rows whose cohort is in {it:numlist}. {phang2} On {cmd:type(dynamic)} and {cmd:type(calendar)} aggregation plots, {cmd:group()} is not meaningful - those aggregations have no cohort dimension left - and {cmd:csdid_plot} says so out loud: {pmore2} {it:group() applies to group-type aggregations; ignored for the dynamic aggregation plot.} {phang2} If any of the requested cohorts does not exist in the results, {cmd:csdid_plot} does not fail and does not produce a partial or empty figure. It reports {pmore2} {it:Some of the specified groups do not exist in the data. Reporting all available groups.} {pmore} names the missing cohorts on a second line, and falls back to every available cohort. {marker remarks}{...} {title:Remarks} {pstd} {cmd:csdid_plot} gives you one picture with no dials, or the plot {it:data} with every dial. That is a deliberate design choice, and it has three consequences worth stating plainly. {phang} {bf:1. The drawn graph takes no styling options.} Options such as {cmd:title()}, {cmd:xtitle()}, {cmd:legend()}, {cmd:name()}, {cmd:style()}, or {cmd:color()} are rejected with {pmore} {it:unsupported option(s): ...} {pmore} and return code 198. The graph {cmd:csdid_plot} draws is a quick look at the estimates in the current scheme. For a figure you control, export with {opt sav:ing()} and pass those options to {helpb twoway} yourself. Nothing about the appearance of a publication csdid figure is baked into this package. {phang} {bf:2. The exported numbers are the numbers.} Every quantity a reader would see in the figure - point estimate, lower bound, upper bound, and whether the interval excludes zero - is a variable in the exported dataset, so a referee, a coauthor, or your future self can check the figure against the estimates without reverse-engineering a graph. {phang} {bf:3. The bounds inherit the inference you asked for.} {cmd:ci_low} is {cmd:estimate - c*se} and {cmd:ci_high} is {cmd:estimate + c*se}, where {cmd:c} is the critical value {cmd:csdid} or {cmd:csdid_stats} already computed at estimation time: {p2colset 5 38 40 2}{...} {p2col :{it:inference requested}}{it:the critical value c}{p_end} {p2line} {p2col :bootstrap with bands (default)}{cmd:e(crit_val)}, the simultaneous (uniform) critical value, or the normal quantile when the band falls back ({cmd:e(cband_fallback)} = 1){p_end} {p2col :bootstrap with {cmd:pointwise}}{cmd:e(crit_val)}, which under {cmd:pointwise} is the normal quantile{p_end} {p2col :{cmd:analytical} (aggregations)}{cmd:e(crit_val)}, the simultaneous critical value the aggregation bootstrapped; with {cmd:pointwise}, or for the ATT(g,t) table, the normal quantile at the reported confidence level{p_end} {p2line} {p2colreset}{...} {pmore} Exactly: the ATT(g,t) plot reads its critical value from {cmd:e(boot_attgt)} (falling back to the normal quantile at {cmd:e(level)} on an analytical run), so an aggregation run in between cannot change the ATT(g,t) band. An aggregation plot uses {cmd:e(crit_val)} whenever the aggregation carried a simultaneous band ({cmd:e(agg_cband)} = 1, under the bootstrap or under {cmd:analytical}), and otherwise the normal quantile at {cmd:e(agg_level)}, the level the aggregation itself reported. {pmore} Either way, the band drawn or exported is the band of the aggregation that is active when {cmd:csdid_plot} runs, at the level that aggregation reported. Because {helpb csdid_stats} and {helpb csdid_estat} recompute the aggregation every time -- bootstrap inference included -- re-running the aggregation at a different {cmd:level()} and plotting again gives a band at the new level, not the old one. {pmore} Because the csdid default is a multiplier bootstrap with simultaneous bands, by default the plotted interval is a {it:uniform} band that covers all reported effects simultaneously with probability {cmd:e(level)}/100 - it is wider than a pointwise 95% interval, and {cmd:significant} therefore marks uniform-band significance, not a pointwise test. If you want pointwise intervals in the figure, estimate with {cmd:pointwise} (see {helpb csdid}), or export {cmd:estat tidy} (see {helpb csdid_estat}), which reports both the band bounds and the pointwise bounds along with standard errors and p-values. {pstd} Standard errors, p-values, the overall aggregated ATT, and the treated/comparison cell counts are {it:not} in the plot dataset; they are one command away in {cmd:estat tidy}, {cmd:e(attgt)}, and {cmd:e(aggte)}. {marker schema}{...} {title:Plot-data schema} {pstd} The dataset written by {opt sav:ing()} always has exactly these eleven variables, in this order. The schema is identical for all plot types, so the same graph code can be reused across figures, and the graph {cmd:csdid_plot} draws for you is built from these same eleven columns. {synoptset 16 tabbed}{...} {synopthdr:variable} {synoptline} {synopt :{cmd:plot_type}}{cmd:str24}; which export this is: {cmd:attgt}, {cmd:aggte_dynamic}, {cmd:aggte_group}, or {cmd:aggte_calendar}{p_end} {synopt :{cmd:series}}{cmd:str8}; {cmd:Pre} or {cmd:Post}; see the note below{p_end} {synopt :{cmd:x}}{cmd:double}; the x-axis value: calendar period ({cmd:attgt}, {cmd:aggte_calendar}), event time ({cmd:aggte_dynamic}), or treatment cohort ({cmd:aggte_group}){p_end} {synopt :{cmd:x_label}}{cmd:str32}; {cmd:x} formatted as text, for use as an axis label{p_end} {synopt :{cmd:estimate}}{cmd:double}; the point estimate: ATT(g,t) or the aggregated ATT{p_end} {synopt :{cmd:ci_low}}{cmd:double}; {cmd:estimate - c*se}{p_end} {synopt :{cmd:ci_high}}{cmd:double}; {cmd:estimate + c*se}{p_end} {synopt :{cmd:group}}{cmd:double}; treatment cohort; filled for {cmd:attgt} and {cmd:aggte_group}, missing otherwise{p_end} {synopt :{cmd:time}}{cmd:double}; calendar period; filled for {cmd:attgt} and {cmd:aggte_calendar}, missing otherwise{p_end} {synopt :{cmd:event_time}}{cmd:double}; length of exposure, t minus g; filled for {cmd:attgt} and {cmd:aggte_dynamic}, missing otherwise{p_end} {synopt :{cmd:significant}}{cmd:byte}; 1 if the interval from {cmd:ci_low} to {cmd:ci_high} excludes zero, 0 otherwise (also 0 when a bound is missing){p_end} {synoptline} {p2colreset}{...} {phang} {bf:series.} On the {cmd:attgt} export, {cmd:series} is {cmd:"Post"} when {cmd:time >= group} and {cmd:"Pre"} otherwise. On aggregation exports it is {cmd:"Post"} when {cmd:x >= 0}. That is exactly right for {cmd:aggte_dynamic}, where {cmd:x} is event time; but note that for {cmd:aggte_group} and {cmd:aggte_calendar} {cmd:x} is a calendar year or cohort, so every row is labeled {cmd:"Post"} and the variable carries no information. Use {cmd:series} for {cmd:attgt} and {cmd:aggte_dynamic} plots; ignore it elsewhere. {phang} {bf:sort order.} The {cmd:attgt} export is sorted by {cmd:group} then {cmd:time}; aggregation exports are sorted by {cmd:x}. {phang} {bf:omitted rows.} The exported rows are exactly the rows of {cmd:e(attgt)} or {cmd:e(aggte)}: whatever window, {cmd:balance()}, or missing-value handling you requested at the aggregation step is already reflected there. Nothing is padded, and no zero-variance placeholder rows are invented for event times that were not estimated. {marker plotnotes}{...} {title:Notes on the exported series} {pstd} Once {opt sav:ing()} has written the dataset, the styling is yours: draw it with {cmd:twoway} and whatever theme, colors, reference line, and {cmd:by(group)} faceting you prefer. The variables it writes, including how {cmd:ci_low}/{cmd:ci_high} and {cmd:series} are built, are documented under {it:{help csdid_plot##schema:Plot-data schema}} above; what follows is the handful of behaviours that are easy to be surprised by, and they apply to the drawn graph as much as to the export. {phang2} o {bf:type(simple) has no plot.} A simple aggregation is a single number, so there is nothing to plot; {cmd:csdid_plot} exits with return code 498 and a message saying so.{p_end} {phang2} o {bf:The reference period is drawn at exactly zero.} Under the default {helpb csdid##opt_base:base_period(universal)} each cohort's base-period cell is the normalisation the other cells are measured against, not an estimate ({it:g-1} on a consecutive calendar without anticipation): it is identically 0 and has no standard error, so it appears as a point on zero with no interval. It is not a precisely estimated null effect, and no pre-trend conclusion should be drawn from it. Under {cmd:base_period(varying)} with no anticipation, the {it:e = -1} cell is an estimated placebo when the required periods exist and does carry an interval. For the universal normalization, the exported dataset shows the same thing: {cmd:estimate} is 0 while {cmd:ci_low} and {cmd:ci_high} are missing.{p_end} {phang2} o {bf:Unknown cohorts} named in {cmd:group()}: when any requested cohort does not exist, it is reported and {cmd:csdid_plot} falls back to every available cohort, so a range such as {cmd:group(2004/2007)} that spans a year with no cohort plots them all.{p_end} {phang2} o {cmd:group()} is honored on {cmd:type(group)} aggregation plots as well as on group-time plots, and {cmd:csdid_plot} reports when it is ignoring it.{p_end} {marker legacy}{...} {title:Relationship to legacy Stata csdid Version 1.82} {pstd} Both versions draw a graph. Legacy {cmd:csdid_plot} also accepted styling options ({cmd:style()}, {cmd:title()}, {cmd:xtitle()}, {cmd:ytitle()}, {cmd:name()}, {cmd:color*}, {cmd:lwidth*}, {cmd:legend()}). Those options are {bf:not accepted} here and are not silently ignored: they raise error 198. Version 2.0.0 draws one standard-scheme graph with no styling surface at all, and hands you the numbers instead when you want a figure of your own. {pstd} So if all you want is the picture, the command is now shorter than it was. Where legacy {cmd:csdid_plot} on plain {cmd:csdid} results answered {it:Please specify group} and drew nothing until you named one, {phang2}{cmd:. csdid_plot}{p_end} {pstd} draws every cohort, and after {cmd:estat event} it draws the event study. {pstd} For styling, the migration is mechanical - export the data, then pass your old styling options to {cmd:twoway}. One thing changes on the way: Version 1.82's {cmd:group()} plot put periods to treatment on the x axis, and this version's cohort panels use the calendar period. The exported data carry both, so to keep the Version 1.82 axis, plot against {cmd:event_time} rather than {cmd:x}. Instead of {phang2}{cmd:. csdid_plot, group(2004) title("Cohort 2004")}{p_end} {pstd} write {phang2}{cmd:. csdid_plot, saving(pd) replace group(2004)}{p_end} {phang2}{cmd:. use pd, clear}{p_end} {phang2}{cmd:. twoway (rcap ci_high ci_low event_time) (scatter estimate event_time), title("Cohort 2004")}{p_end} {pstd} {cmd:group()} keeps its legacy meaning (select treatment cohorts) and, unlike in some legacy paths, it is now honored on group-type aggregation plots rather than silently discarded. See {help csdid_postestimation} and {help csdid_legacy}. {marker examples}{...} {title:Examples} {pstd}{bf:Setup: a small staggered-adoption panel}{p_end} {phang2}{cmd:. clear}{p_end} {phang2}{cmd:. set seed 12345}{p_end} {phang2}{cmd:. set obs 2000}{p_end} {phang2}{cmd:. generate long id = ceil(_n/5)}{p_end} {phang2}{cmd:. bysort id: generate int year = 2003 + _n}{p_end} {phang2}{cmd:. generate int gvar = cond(id <= 160, 0, cond(id <= 280, 2005, 2007))}{p_end} {phang2}{cmd:. generate byte treated = gvar > 0 & year >= gvar}{p_end} {phang2}{cmd:. generate double x = rnormal()}{p_end} {phang2}{cmd:. generate double y = .3*(year - 2003) + .5*treated}{p_end} {phang2}{cmd:. replace y = y + .2*x + rnormal()}{p_end} {pstd}{bf:Estimate, then draw the ATT(g,t) plot: one panel per cohort}{p_end} {phang2}{cmd:. csdid y x, ivar(id) time(year) gvar(gvar) ///}{break} {cmd: wboot(reps(200) rseed(20260726))}{p_end} {phang2}{cmd:. csdid_plot}{p_end} {pstd}{bf:Draw the event study}{p_end} {phang2}{cmd:. estat event}{p_end} {phang2}{cmd:. csdid_plot}{p_end} {pstd}{bf:Draw one cohort only}{p_end} {phang2}{cmd:. csdid y x, ivar(id) time(year) gvar(gvar)}{p_end} {phang2}{cmd:. csdid_plot, group(2007)}{p_end} {pstd}{bf:Export ATT(g,t) plot data instead of drawing}{p_end} {phang2}{cmd:. csdid y x, ivar(id) time(year) gvar(gvar) ///}{break} {cmd: wboot(reps(200) rseed(20260726))}{p_end} {phang2}{cmd:. csdid_plot, saving(attgt_plotdata) replace}{p_end} {pstd}{bf:Export event-study (dynamic aggregation) plot data}{p_end} {phang2}{cmd:. csdid_stats event, dropmissing}{p_end} {phang2}{cmd:. csdid_plot, saving(event_plotdata) replace}{p_end} {pstd}{bf:Export cohort (group aggregation) plot data for one cohort only}{p_end} {phang2}{cmd:. csdid_stats group}{p_end} {phang2}{cmd:. csdid_plot, saving(group_plotdata) replace group(2007)}{p_end} {pstd}{bf:Inspect what was written}{p_end} {phang2}{cmd:. use attgt_plotdata, clear}{p_end} {phang2}{cmd:. describe}{p_end} {phang2}{cmd:. list in 1/6, clean noobs}{p_end} {pstd}{bf:Style your own group-time figure from the exported data}{p_end} {phang2}{cmd:. use attgt_plotdata, clear}{p_end} {phang2}{cmd:. twoway (rcap ci_high ci_low x if series == "Pre", lcolor(gs9)) ///}{break} {cmd: (scatter estimate x if series == "Pre", mcolor(gs9)) ///}{break} {cmd: (rcap ci_high ci_low x if series == "Post", lcolor(navy)) ///}{break} {cmd: (scatter estimate x if series == "Post", mcolor(navy)), ///}{break} {cmd: by(group, note("")) yline(0) xtitle("Calendar period") ///}{break} {cmd: ytitle("ATT(g,t)") ///}{break} {cmd: legend(order(2 "Pre-treatment" 4 "Post-treatment"))}{p_end} {pstd}{bf:Style your own event study, with a shaded band}{p_end} {phang2}{cmd:. use event_plotdata, clear}{p_end} {phang2}{cmd:. twoway (rarea ci_high ci_low x, color(navy%20)) ///}{break} {cmd: (connected estimate x, mcolor(navy) lcolor(navy)), ///}{break} {cmd: yline(0) xline(-0.5) xtitle("Event time") ytitle("ATT(e)") ///}{break} {cmd: legend(off)}{p_end} {pstd} The do-files in the {cmd:examples/} directory at {browse "https://github.com/pedrohcgs/csdid-stata"} run the same workflow end to end, including on the Callaway and Sant'Anna county teen-employment data ({cmd:examples/06_mpdta_workflow.do}). They are not part of the net-install package. {marker results}{...} {title:Stored results} {pstd} {cmd:csdid_plot} stores nothing. It is not an {cmd:eclass} command: the estimation results left by {cmd:csdid}, {cmd:csdid_stats}, or {cmd:estat} are still in {cmd:e()} afterwards, unchanged, so you can draw several figures or export several plot datasets in a row and then continue with other postestimation commands. The data in memory are also unchanged. {pstd} Its output is a graph, or, under {opt sav:ing()}, the dataset named there, documented in {help csdid_plot##schema:Plot-data schema} above. {marker errors}{...} {title:Diagnostics} {p2colset 5 12 14 2}{...} {p2col :{it:rc}}{it:condition and message}{p_end} {p2line} {p2col :{cmd:301}}No {cmd:csdid} results in {cmd:e()}: {it:csdid_plot requires prior csdid results}{p_end} {p2col :{cmd:198}}An option that is not {cmd:saving()}, {cmd:replace}, or {cmd:group()} was given: {it:unsupported option(s): ...}{p_end} {p2col :{cmd:198}}Something other than {cmd:replace} follows the comma in {cmd:saving()}: {it:saving() accepts only the replace sub-option; cannot parse: ...}{p_end} {p2col :{cmd:198}}{cmd:saving()} has a comma but no filename before it: {it:saving() requires a filename before the comma, as in saving(myfile, replace)}{p_end} {p2col :{cmd:198}}{cmd:replace} was given without {cmd:saving()}: {it:replace has no effect without saving(); specify saving(filename) or drop replace}{p_end} {p2col :{cmd:498}}The active aggregation is {cmd:type(simple)}, which is one overall effect and so has no axis to plot against: {it:Plot method not available for this type of aggregation}. Plot a {cmd:dynamic}, {cmd:group}, or {cmd:calendar} aggregation instead, or re-run {cmd:csdid} to plot the ATT(g,t) estimates{p_end} {p2col :{cmd:498}}Every estimate to be drawn is missing: {it:nothing to plot: every estimate is missing}, followed by the causes to check in the {cmd:csdid} run that produced them{p_end} {p2col :{cmd:602}}{it:filename} exists and {cmd:replace} was not specified{p_end} {p2line} {p2colreset}{...} {pstd} Two messages are notes, not errors: the {cmd:group()}-is-ignored note on dynamic and calendar aggregation plots, and the fall-back-to-all-cohorts note when a requested cohort does not exist. Both are described under {help csdid_plot##opt_group:group()}. {pstd} A successful {opt sav:ing()} export prints nothing and draws nothing. The scratch-dataset work {cmd:csdid_plot} does to build the file is silent, so neither observation-count notices nor a {cmd:more} prompt interrupts the export; the notes above are the only things it ever says on a run that succeeds. A successful bare {cmd:csdid_plot} likewise prints no text: its output is the graph. {marker references}{...} {title:References} {marker baker2026}{...} {phang} Baker, A., B. Callaway, S. Cunningham, A. Goodman-Bacon, and P. H. C. Sant'Anna. 2026. Difference-in-differences designs: A practitioner's guide. {it:Journal of Economic Literature} 64(2): 498-557. {browse "https://doi.org/10.1257/jel.20251650"}. {p_end} {marker callaway2021}{...} {phang} Callaway, B., and P. H. C. Sant'Anna. 2021. Difference-in-differences with multiple time periods. {it:Journal of Econometrics} 225(2): 200-230. {browse "https://doi.org/10.1016/j.jeconom.2020.12.001"}. {p_end} {marker roth2023}{...} {phang} Roth, J., P. H. C. Sant'Anna, A. Bilinski, and J. Poe. 2023. What's trending in difference-in-differences? A synthesis of the recent econometrics literature. {it:Journal of Econometrics} 235(2): 2218-2244. {browse "https://doi.org/10.1016/j.jeconom.2023.03.008"}. {p_end} {marker authors}{...} {title:Authors} {pstd} {cmd:csdid} is by Brantly Callaway, Fernando Rios-Avila, and Pedro H. C. Sant'Anna. Full affiliations and contact addresses, the acknowledgments, how to report a problem, and how to cite the package are in {help csdid##authors:help csdid}. Please report anything that looks numerically wrong with a reproducible example - the plot data are derived from {cmd:e(attgt)} and {cmd:e(aggte)}, so include the {cmd:csdid} command line that produced them. {title:Also see} {psee} Online: {helpb csdid}, {helpb csdid_postestimation:csdid postestimation}, {helpb csdid_stats}, {helpb csdid_estat}, {helpb twoway} {p_end}