Reading Regression Tables

The text below is a transcript of the video above, edited with AI assistance for clarity and flow. Where the two differ, the video is the authoritative version.

You’ll learn to

Whenever you encounter a regression table in a quantitative social science publication, you will see it laid out in a way that is very different from what you see when you conduct your own analysis using software. In this video you will learn to:

  • Identify the parts of a publication-style regression table.
  • Read coefficients, standard errors, and significance stars. The stars are a new element here.
  • Compare multiple models presented side by side.

Regressions “in the wild”

In the wild, when you are reading papers, reports and books, you will find regression tables that look nothing like the output you see when you run a regression yourself in whatever software you are using. Regression tables in publications show you multiple regressions in a single table. This video teaches you how to read them.

The full table

Dependent variable:
Leave Vote Share (%)
(1) (2) (3)
Higher Education (%) −1.130***(0.147) −1.143***(0.148)
Immigrant Share (%) −0.414***(0.066) 0.085(0.112)
Constant 83.918***(3.995) 56.807***(1.110) 83.493***(4.040)
Observations 139 167 139
R2 0.301 0.193 0.304
Adjusted R2 0.296 0.188 0.294
Note: *p<0.1; **p<0.05; ***p<0.01

This is the full table. The dependent variable is the Leave vote share, and the table contains three regressions that you should be familiar with at this point. Number one is a bivariate regression of the Leave vote on higher education. Number two, in the middle, is a bivariate regression of the Leave vote on immigrant share. Number three is a multivariate regression that contains both higher education and immigrant share on the right-hand side. So you have three different regression models in one table.

Each column is a model

Dependent variable:
Leave Vote Share (%)
(1) (2) (3)
Higher Education (%) −1.130***(0.147) −1.143***(0.148)
Immigrant Share (%) −0.414***(0.066) 0.085(0.112)
Constant 83.918***(3.995) 56.807***(1.110) 83.493***(4.040)
Observations 139 167 139
R2 0.301 0.193 0.304
Adjusted R2 0.296 0.188 0.294
Note: *p<0.1; **p<0.05; ***p<0.01

Each column is a model. In the first column, the higher education model, you have the intercept of about 83.9 and the slope of −1.13. In the second, the bivariate immigrant share model, you have the intercept of about 56.8 and the slope of −0.41. And in the third column, model number three, you have the intercept and both slopes. Each column is a different regression model.

The dependent variable

Dependent variable:
Leave Vote Share (%)
(1) (2) (3)
Higher Education (%) −1.130***(0.147) −1.143***(0.148)
Immigrant Share (%) −0.414***(0.066) 0.085(0.112)
Constant 83.918***(3.995) 56.807***(1.110) 83.493***(4.040)
Observations 139 167 139
R2 0.301 0.193 0.304
Adjusted R2 0.296 0.188 0.294
Note: *p<0.1; **p<0.05; ***p<0.01

The dependent variable is usually specified at the top. Here it says Leave Vote Share. That is where the dependent variable is displayed.

Each row is a variable

Dependent variable:
Leave Vote Share (%)
(1) (2) (3)
Higher Education (%) −1.130***(0.147) −1.143***(0.148)
Immigrant Share (%) −0.414***(0.066) 0.085(0.112)
Constant 83.918***(3.995) 56.807***(1.110) 83.493***(4.040)
Observations 139 167 139
R2 0.301 0.193 0.304
Adjusted R2 0.296 0.188 0.294
Note: *p<0.1; **p<0.05; ***p<0.01

Each row is a variable. So you have models in columns, variables in rows, and the specification of the dependent variable at the top.

Blank cells

Dependent variable:
Leave Vote Share (%)
(1) (2) (3)
Higher Education (%) −1.130***(0.147) −1.143***(0.148)
Immigrant Share (%) −0.414***(0.066) 0.085(0.112)
Constant 83.918***(3.995) 56.807***(1.110) 83.493***(4.040)
Observations 139 167 139
R2 0.301 0.193 0.304
Adjusted R2 0.296 0.188 0.294
Note: *p<0.1; **p<0.05; ***p<0.01

A blank cell means that the variable was not included in that model. In model one, for example, immigrant share was not included, so you get a blank cell: no coefficient there. Model two does not include higher education as a variable, so there is a blank cell there too. There will be some blank cells in most regression tables that you see, because there will be models that do not include certain variables.

Stars indicate significance

Dependent variable:
Leave Vote Share (%)
(1) (2) (3)
Higher Education (%) −1.130***(0.147) −1.143***(0.148)
Immigrant Share (%) −0.414***(0.066) 0.085(0.112)
Constant 83.918***(3.995) 56.807***(1.110) 83.493***(4.040)
Observations 139 167 139
R2 0.301 0.193 0.304
Adjusted R2 0.296 0.188 0.294
Note: *p<0.1; **p<0.05; ***p<0.01

The stars are visual shortcuts that these tables use to indicate statistical significance. At the bottom of the table there is a note indicating what each number of stars means for the p-value. You do not get the exact p-value, and you do not get the full confidence interval. You get stars that tell you which p-value threshold the estimate meets.

In this case:

  • One star means the p-value is smaller than 0.1.
  • Two stars mean it is smaller than 0.05.
  • Three stars mean it is smaller than 0.01.

This gives you a sense of the statistical significance of each estimate. When you get no stars, in this case it means that the p-value is larger than 0.1.

Standard errors

Dependent variable:
Leave Vote Share (%)
(1) (2) (3)
Higher Education (%) −1.130***(0.147) −1.143***(0.148)
Immigrant Share (%) −0.414***(0.066) 0.085(0.112)
Constant 83.918***(3.995) 56.807***(1.110) 83.493***(4.040)
Observations 139 167 139
R2 0.301 0.193 0.304
Adjusted R2 0.296 0.188 0.294
Note: *p<0.1; **p<0.05; ***p<0.01

The standard errors are right below the coefficient, in parentheses. So you have the estimate, with stars if it is significant, and right below it the standard error.

The bottom panel

Dependent variable:
Leave Vote Share (%)
(1) (2) (3)
Higher Education (%) −1.130***(0.147) −1.143***(0.148)
Immigrant Share (%) −0.414***(0.066) 0.085(0.112)
Constant 83.918***(3.995) 56.807***(1.110) 83.493***(4.040)
Observations 139 167 139
R2 0.301 0.193 0.304
Adjusted R2 0.296 0.188 0.294
Note: *p<0.1; **p<0.05; ***p<0.01

The bottom panel contains additional information about each model. In this case you have the number of observations, the R-squared, and the adjusted R-squared for each of the three regression models.

That is the overall look of a publication-style regression table. What is important is that you know where to find each of the key estimates when you encounter these tables in publications.

Recap

  • Each column is a separate regression model.
  • Stars indicate significance; the numbers in parentheses are the standard errors.
  • Comparing columns shows how the coefficients change across models.

Why this matters

Published research presents results in this format, not as software output. Reading regression tables is very important for being able to engage with quantitative social science research.