Residual Plots in Tableau - why are they so important?

by Elnisa Marques

Sometimes when using a Linear Regression to analyse our data, we may think that the model we are using is the best fit for our data because the p-value is significant (p-value< 0.05) and the r-square value is close to 1 (R² is high), however this may not always be the case. We should always assess the appropriateness of the model by defining the residuals and examining the residual plots.

residual plot4

But what is a residual?

A residual is the difference between the observed value of the dependent variable (y) and the predicted value (ŷ). Each data point has one residual.

Residual = Observed value – Predicted value
                                e = y – ŷ

Note: The sum and the mean of the residuals are equal to zero.

residual plot

And a residual plot, what is it?

residual plot is a graph that shows the residuals on the vertical axis and the independent variable on the horizontal axis. If the points in a residual plot are randomly dispersed around the horizontal axis, this means that our linear regression model is appropriate for the data; otherwise, a non-linear model is more appropriate.

The residual plots show ‘two typical’ patterns: a random pattern (indicating a good fit for a linear model) and a non-random pattern (U-shaped and inverted U), suggesting a better fit for a non-linear model.

residuals1

 

How do we do a residual plot in tableau?

1 – On the sheet that you have visualised your scatterplot, go to worksheet menu and select export data

residual plot3 residual plot6

 

2 – On the menu box select ‘Connect after Export’ and click OK (this will now be saved as an Access file and it will open automatically in Tableau)

residual plot..

 

 

3- On a new sheet, drag the recently created field ‘residuals’ to the rows and your independent variable to columns (x axis) – in this example ‘wind speed’

residuals plot

  • We can now see that the residuals aren’t randomly distributed, they do follow a pattern – a ‘sigmoid distribution’. This tell us, that contrary of what we believed (by analysis of p-value and r-square) the Linear Regression model isn’t the best fit model to our data. We should now look for other types of models, namely the non-linear models.

Example of a residual plot showing that our linear regression model is the best fit to our data                                                                                                                                                                              residual plot9

 

 

 

 

 

 Random Distribution around zero

 

Note: when accessing p-value and R-squared don’t forget to analyse the number of observations and degrees of freedom, as these may indicate an artificially high r-squared value.

High number of observations and low degrees of freedom indicate that the high r-square value may be due to external reasons.

residual plot5