Interpret regression output in plain language

Paste a regression table and get a careful interpretation of coefficients, fit and reliability, with warnings about causal claims.

LLazyPrompt(@lazyprompt)Data & AnalysisAnalyzeWorks with any model
No ratings yetSign in to rate
Save 0 Fork Report
Prompt template
Interpret this regression output for me in plain language, and check whether the model is trustworthy.

Model type: {{model_type}}
What the model tries to predict or explain: {{outcome}}
Predictors and how they are measured (units, coding): {{predictors}}
Output (paste the table or summary):
{{output}}
Why I ran this: {{purpose}}

Provide:
1. Model in one sentence: what it estimates, in simple words.
2. Coefficient by coefficient: the meaning in real units ("holding the other variables constant, one more X is associated with Y change in the outcome"), the direction and size, and whether the uncertainty (confidence interval or standard error) leaves the effect clearly different from zero. For logistic or log-transformed models, convert to odds ratios or percentage changes and show the arithmetic.
3. Fit and explanatory power: what R-squared or the equivalent says and does not say.
4. Diagnostics I should check next: multicollinearity, non-linearity, outliers, heteroscedasticity, missing data, and overfitting, with the specific plot or test for each and what a problem looks like. Mention any warning signs visible in the output I pasted.
5. Causal caution: what this model can and cannot claim given how the data was collected, and possible confounders.
6. A 3-sentence statement I could present to a non-technical audience that does not overclaim.
7. Suggested improvements or a next model.

If the output is incomplete or ambiguous, say what is missing instead of guessing.
Example output
**Model**: A linear model estimating how sale price (in $ thousands) changes with size, bedrooms, age and distance from downtown, using 480 sales. **Coefficients** - sqft 0.148: each extra square foot is associated with $148 higher price, holding the rest constant. 100 more sqft is about $14.8k. Very precise (SE 0.011). - bedrooms -6.2: one more bedroom is associated with $6.2k lower price at the same size. The standard error (4.9) is large relative to the estimate, so this could easily be zero. It is negative because size is already in the model; more bedrooms in the same space means smaller rooms. Do not read it as "bedrooms reduce value". - age -1.1: each extra year, about $1.1k lower. Clear effect. - distance -9.8: each extra mile from downtown, about $9.8k lower. Clear effect. - Intercept 52.3 is a price for a zero-size, zero-everything house; it has no meaning. **Fit**: R-squared 0.71 means the model explains about 71% of the variation in this sample. It does not show the model predicts new houses well. **Check next**: bedrooms and sqft correlate (VIF); residuals vs fitted for fan shapes (price variance often grows with price, consider log price); outlier mansions (Cook's distance); hold-out validation. **Causal caution**: These are associations. Neighbourhood quality, condition and school district are omitted and may drive both distance and price. **For a presentation**: In our sample, bigger, newer houses closer to downtown sell for more. Size matters most. We cannot conclude that changing any one factor would change a price.

Why it works

Translating each coefficient into real units and flagging what the model cannot claim prevents the usual over-interpretation. For better diagnostics, add the residual plots described or summary statistics for each variable.

Licensed CC0 (public domain) Version 1 · history