What it does
This is a regression calculator for economics and the social sciences that runs entirely in your browser. Open a CSV or paste a table, pick the outcome and the explanatory variables, and choose one of three estimators:
- OLS (ordinary least squares), with classical standard errors, heteroskedasticity-robust standard errors (HC0, HC1, HC2 or HC3), or standard errors clustered by any column such as school, firm or region.
- IV (2SLS): two-stage least squares for a regressor that is endogenous, using one or more excluded instruments. The tool also shows the first-stage F and partial R² for each endogenous variable, so you can see whether the instruments are strong.
- Panel fixed effects: the within estimator, with entity effects and, if you tick it, time effects as well (two-way fixed effects). It works with unbalanced panels.
You get a coefficient table with standard errors, t statistics, p-values and confidence intervals, plus R², adjusted R², a Wald F test of all slopes, the residual standard error, and a plot of residuals against fitted values. Download the report as an image, PDF or Word file, or the coefficients as CSV.
It works well with the publication plot maker, the trend and anomaly analyser and the sample size and power calculator.
How to use it
- Paste your data or use Open a file. The first row must hold the column names. Each row is one observation; for a panel, one row per entity and period.
- Pick OLS, IV (2SLS) or Panel fixed effects.
- Choose the outcome and tick the explanatory variables. For IV, also tick the endogenous variables and the excluded instruments (the instruments must not appear among the explanatory variables).
- Choose the standard errors. If you choose clustered, pick the column with the groups. For panels, pick the entity and time columns.
- Rows with a blank or non-numeric cell in any variable you use are left out, and the tool says how many.
The options in one line each
- Classical: assumes every error has the same spread and errors are independent.
- HC0: White’s robust standard errors, valid when the spread differs between observations.
- HC1: HC0 times n/(n − k), the same as Stata’s “robust”.
- HC2 and HC3: adjust each residual for its leverage; HC3 is the safer choice in small samples.
- Clustered: allows errors to be correlated inside a group. For OLS, tests use G − 1 degrees of freedom, where G is the number of clusters.
- Time effects: adds a dummy for each period, removing shocks that hit every entity in the same year.
Worked example (made-up data)
The sample is a made-up panel of 15 people over 4 years, with log wage, years of schooling (educ), experience, union membership and the distance to the nearest college when they were young.
OLS: log_wage on educ, exper and union gives educ = 0.0582 (SE 0.0145, p = 0.0002), exper = 0.0426 and union = 0.0593 (p = 0.29), with R² = 0.3128 and F(3, 56) = 8.50. With HC3 robust errors, the SE of educ becomes 0.0170.
IV: schooling is likely endogenous, because unmeasured ability raises both schooling and wages. Using distance as the instrument for educ, the first-stage F is 204.4 (strong), and the 2SLS estimate of educ falls to 0.0292 (SE 0.0170, p = 0.09). The OLS estimate was pushed up by ability.
Fixed effects: with person effects, exper = 0.0473 (SE 0.0105) and union = 0.0739 (SE 0.0431), within R² = 0.3360. Schooling is left out with a note, because it never changes within a person, so the person effects absorb it.
How it was checked
Every number is checked against statsmodels (OLS, all five covariance types, with t-based inference) and linearmodels (IV2SLS with debiased=True, and PanelOLS with its defaults), on the example data and on an unbalanced panel. They agree to at least nine significant figures.
Frequently asked questions
Which robust option should I use? HC1 is the most widely reported; HC3 is better below about 250 observations. Clustering is needed when observations in the same group share shocks.
Can I use text categories? Explanatory variables must be numbers. Make 0/1 dummy columns first (the data cleaner can help). Text columns can still be the cluster, entity or time column.
Why can R² be negative for IV? 2SLS does not minimise the residual sum of squares, so R² is not a measure of fit there. Judge an IV model by its first stage and its reasoning instead.
Is my data uploaded? No. Everything runs in your browser and the page keeps working offline.
Where the numbers come from
- Formulas: White (1980) and MacKinnon and White (1985) for HC0 to HC3; Liang and Zeger (1986) for clustered errors; two-stage least squares and the within estimator as in Wooldridge, Econometric Analysis of Cross Section and Panel Data. Example data made up.
Standard scientific values, such as constants, atomic weights and melting points, are facts found in every textbook; we name where ours were checked. If you are an author or publisher and want a value, a credit or a table changed or removed, tell us and we will act quickly.
Last updated 2026-11-09.