4

I fitted this lmer model:

m1 <- lmer(logR ~ N_g.m.2 * Year + (1|Wh/N_g.m.2), data = CO2_Ratio)

Rendering the attached qqplot.

qqnorm(resid(m1))
qqline(resid(m1))

What does it mean?enter image description here

fede_luppi
  • 256
  • 1
  • 4
  • 17

1 Answers1

3

The Q-Q plot is a probability plot of the standardized residuals against the values that would be expected under normality. If the model residuals are normally distributed then the points on this graph should fall on the straight line, if they don't, then you have violated the normality assumption.

Your plot shows that your model does does badly on low and high "outlier" values -- this was what I originally wrote, but consider also EdM's comments below; they supplements this answer, expand and offer more insight.

  • Don't know that I'd say that the model "does badly on low and high 'outlier' values." The model doesn't lead to a normal distribution of residual values, but that doesn't necessarily mean that the model is bad. It just means that any further tests that depend heavily on normally distributed residuals might be called into question. The model itself could be quite useful. It's much more concerning if the plot of residuals against predicted values shows poor behavior. – EdM Mar 21 '17 at 20:38
  • Your are right, but I would question the model's predictions and I would probably exclude model outliers and refit the model to check if outliers influence estimates. –  Mar 21 '17 at 21:00
  • If the scales on the plot are correct, this would seem to be a "light-tailed" distribution compared to a normal distribution (see [this helpful answer](http://stats.stackexchange.com/a/101290/28500) for a gallery of QQ-plots). The observed residuals are more centrally located than you would expect from a normal distribution. (The line on the plot is not the 45-degree line.) So we're talking more about in-liers than out-liers here. – EdM Mar 21 '17 at 21:56