Research & PolicyFree
Regression Results Section From Your Output
River writes the results section from the output block you paste, and names every statistic the standard asks for that your software never printed.
River reads the output block you paste and writes the results section from it, with every figure traced to the line it came from. A statistic your software never printed is named as a gap rather than quietly supplied. That distinction is most of the job. R's reference page for summary describes the coefficient table as a matrix of the estimate, its standard error, the t statistic and the p value. Four columns, so a confidence interval in a write-up from that paste was derived, and the draft says so.
The transformation is different in every package, which is why a general checklist never finishes the job. Stata prints the interval by default, and its own manual says the beta option reports standardized coefficients instead of confidence intervals. Instead of, not alongside. APA's reporting standard meanwhile asks for effect sizes and confidence intervals together, for every inferential test conducted. One command cannot satisfy that. River names the second call that supplies the missing column, then writes the sentence.
It will not report an assumption nobody tested. If the paste contains no normality test, the draft says the check is absent rather than describing residuals as approximately normal. The same rule governs the sample size, the degrees of freedom and every effect size: present in the output, or named as missing. The analytic N has to come from somewhere too, and the cleaning record is where it derives from the raw N by arithmetic. A matcher that documents its own blind spot beats one reporting a clean pass.
One real model, and four numbers that do not survive the copy-paste
Fit ozone on solar radiation, wind and maximum temperature in R's airquality data, the 153 rows that ship with base R. Temperature's p value is 2.42e-09 and wind's is 1.52e-06. At three decimals both display as 0.000, and neither is zero. APA's floor is p < .001, so both are written that way. The inverse error costs as much: solar radiation's p is .011, an exact value that belongs in the sentence as .011 rather than rounded up to a threshold.
Ranked by raw coefficient, wind leads at 3.33 and temperature follows at 1.65. Ranked by standardized beta, temperature leads at .47 and wind follows at .36. The top predictor changes, and the only reason wind's coefficient is 2.02 times larger is that miles per hour and degrees Fahrenheit are different units. R's default block carries neither the beta nor the interval, so both were computed after the fact from the standard deviations. A write-up that does not say so is claiming output it never had.
The sample size moves under you. R fitted 111 of those 153 days and printed the reason, that 42 observations were deleted due to missingness, because its factory-fresh default drops incomplete rows. Drop solar radiation and the sample becomes 116, so comparing the two R-squared values gives .037. On the same 111 days it is .025. The naive figure overstates the gain by half, and the nested test that settles it exists only on the shared rows: F(1, 107) = 6.66, p = .011.
How it works
Paste the output
The coefficient table as your software printed it, with the fit lines underneath, and the package it came from.
Columns get named
Each column is matched to the statistic it holds, which differs by package even for the same model.
Gaps get flagged
What the standard requires and the output does not contain is listed, with the call that would supply it.
The section is written
Prose in the standard's own conventions, with every number carrying the line of output it was read from.
What you get
- The results section written from your output block, with each figure traced to its line
- A coefficient table as a Sheet: B, the standard error, the interval, beta and exact p
- Every statistic the standard asks for that your output never contained, named as absent
- The sample size the model actually used, and how many rows the software dropped
- Assumption checks reported only where the output carries the statistic that tests them
- A chart of the coefficients with their intervals, so precision is visible not asserted
Common questions
Why does my p value show as .000?
Because the display rounded it, not because it is zero. Three decimals is a display width, and Stata's manual says plainly that its own 0.000 means the significance is below 0.0005. The value underneath is still a number: in the worked example on this page it is 2.42e-09. APA asks you to write p < .001, which is a different statement from p = 0.
Will it report my assumption checks?
Only the ones your output actually contains. If the paste carries a Shapiro-Wilk statistic it gets reported, including when it fails. The model in the worked example has two predictors under p < .001 and non-normal residuals, and the honest sentence says both. If the paste carries no test for an assumption, the draft names the absence rather than describing residuals it never saw.
Do I need to upload a file?
No. Paste the block. A results section is built from what the output says, and the text of a coefficient table carries all of it. A console paste and a plain-text log both work. Proprietary viewer formats are a separate problem this does not claim to solve, so name the package that printed the block and the columns get read accordingly.
Which standard does it write to?
The one you pick. APA's own guide asks for exact p values to two or three decimals with a floor at .001, and a journal's author guide can ask for something different. The draft follows the one you name rather than blending several into a house style no venue asked for. If you are not sure yet, it writes to APA and says so at the top.
Can it compare two models for me?
It can, and it checks the sample first. Two R-squared values are only comparable when both were computed on the same rows, and listwise deletion moves that quietly. In the worked example the naive difference is .037 and the honest one is .025. Where the samples differ it says so, the way two reports that will not agree have to be handled.
I have interview data too. Does this cover it?
No, and it should not pretend to. Coding transcripts into themes is a different operation with different failure modes, and synthesizing interviews is where that belongs. This page is for a fitted model's output block. If your paper carries both, the quantitative section is written here and the qualitative one there, which keeps each set of conventions intact.
Does it decide what my results mean?
No. It writes what the model estimated, in the standard's conventions, and stops where interpretation starts. A coefficient on observational data is an association held at fixed values of the other predictors, and the draft phrases it that way rather than as a cause. What the finding means for your paper is yours, as it is for an experiment reported with its interval.
Regression Results Section From Your Output
Fill in the form and your workspace opens with the work already underway.