Post-estimation projection and boundary analysis for Stata
regproject is a post-estimation command that projects the effect of a chosen independent variable (IV) on the dependent variable (DV), while holding all other regressors at their median values. It automatically detects whether the active dataset is cross-sectional, time-series, or panel, and produces a mode-appropriate set of publication-quality graphs.
ssc install regprojectnet install regproject, from("https://raw.githubusercontent.com/nomanarshed/regproject/main/") replaceOr manually: download all .ado and .sthlp files and place them in your personal ado directory (adopath).
regproject varname [, ivmin(#) ivmax(#)
ivmins(numlist) ivmaxs(numlist)
ymin(#) ymax(#)
saving(stub) combine nodisplay]| Option | Description |
|---|---|
varname |
Focal independent variable to project |
ivmin(#) / ivmax(#) |
Lower / upper bound for focal IV only |
ivmins(numlist) / ivmaxs(numlist) |
Bounds for all IVs in regression order (positional) |
ymin(#) / ymax(#) |
DV limits for shading and crossing detection |
saving(stub) |
Save graphs as stub_1.gph, stub_2.gph, ... |
combine |
Combine all mode graphs into a single 2×2 figure |
nodisplay |
Suppress screen display (useful in batch scripts) |
Note: ivmin/ivmax and ivmins/ivmaxs are mutually exclusive. Use the scalar form for the focal variable only; use the list form to set bounds for every regressor simultaneously.
regproject reads xtset and tsset status automatically:
| Condition | Mode detected | Graphs produced |
|---|---|---|
No xtset / tsset active |
Cross-sectional | 4 graphs |
tsset or xtset with time variable only |
Time series | 3 graphs |
xtset with panel and time variable |
Panel data | 4 graphs |
| Graph | Type | What it shows |
|---|---|---|
| 1 | Bar chart | Predicted ŷ per entity using own covariate values, sorted by focal IV; four horizontal reference lines; DV shading |
| 2 | Horizontal bar | Sensitivity ranking — all IVs ranked by |coef × (upper − lower)|; focal IV highlighted |
| 3 | Diverging bar | Gap-to-boundary — headroom to DV upper limit or depth below DV lower limit per entity |
| 4 | Grouped bar | Actual ŷ vs counterfactual ŷ (focal IV replaced with its median) |
| Graph | Type | What it shows |
|---|---|---|
| 1 | Dot chart | Predicted ŷ per entity at latest observed period, sorted by focal IV; four reference lines; DV shading |
| 2 | Spaghetti | Entity ŷ trajectories over time; median trajectory in red |
| 3 | Scatter | Δ focal IV (first to last period) vs Δ ŷ per entity; theoretical regression slope as dashed reference |
| 4 | Heat map | Entities × time periods, cell colour = predicted ŷ (requires heatplot from SSC; fallback bubble scatter otherwise) |
| Graph | Type | What it shows |
|---|---|---|
| 1 | Line + ribbon | Static sweep of focal IV from lower to upper bound; 95% CI ribbon; red diamond at latest observed value |
| 2 | Line | 25-period forward projection; focal IV trended linearly, others held at median; DV crossing annotated |
| 3 | Line | 25-period full-system projection; all covariates trended simultaneously; inhibitor dynamics reflected |
Every call prints a table to the results window before any graph is drawn, documenting:
- Detected data mode and model command
- DV bounds and their source
- Every regressor with its coefficient, lower bound, upper bound, and source label (
data range/ivmin/ivmax/ivmins/ivmaxs)
This provides a reproducible audit trail of all assumptions embedded in the graphs.
sysuse auto, clear
regress price mpg weight length
regproject weight, ivmins(10 1500 140) ivmaxs(41 5000 233) ymin(3000) ymax(16000) combinesysuse uslifeexp, clear
tsset year
regress le_wmale le_wfemale le_bmale
regproject le_wfemale, ivmin(55) ivmax(95) ymin(55) ymax(85) combine* Generate a 10-country 5-period panel
clear
set obs 50
generate cid = ceil(_n / 5)
generate year = 2010 + mod(_n-1, 5) * 2
generate str15 cname = ""
replace cname = "Malaysia" if cid==1
replace cname = "Thailand" if cid==2
replace cname = "Indonesia" if cid==3
replace cname = "Philippines" if cid==4
replace cname = "Vietnam" if cid==5
replace cname = "Cambodia" if cid==6
replace cname = "Bangladesh" if cid==7
replace cname = "Pakistan" if cid==8
replace cname = "SriLanka" if cid==9
replace cname = "India" if cid==10
generate gdp = 3 + 4*runiform() + 0.2*(year-2010)/2
generate trade = 40 + 80*runiform()
generate govexp = 15 + 10*runiform()
generate poverty = 35 - 1.8*gdp - 0.05*trade - 0.3*govexp + rnormal(0,1.5)
xtset cid year
xtreg poverty gdp trade govexp, fe
regproject gdp, ivmins(-2 25 10) ivmaxs(10 150 35) ymin(0) ymax(40) combineA complete self-contained example do-file covering all three modes is included in the package as regproject_example.do.
regress, xtreg, qreg, iqreg, xtregar, newey, arima, ardl, ivregress
Any command that stores results in e(b), e(V), and e(depvar) and supports predict, xb should work in practice.
heatplot (Jann, SSC) is used for Panel Graph 4. If not installed, regproject falls back to a bubble scatter automatically.
ssc install heatplotIf you use regproject in published work, please cite:
Arshed, N. (2026). regproject: Post-estimation projection and boundary analysis for Stata. Statistical Software Components, Boston College Department of Economics.
Dr Noman Arshed
Senior Lecturer, Department of Business Analytics
Sunway Business School, Sunway University
📧 nouman.arshed@gmail.com
🐙 github.com/nomanarshed/regproject
Bug reports and feature requests are welcome via GitHub Issues.
| Version | Date | Notes |
|---|---|---|
| 1.0.0 | April 2026 | Initial release |