Are there regression models where variance is the outcome, not mean? For instance, for interquartile range I may use quantile regression. But is there something similar for variance?

For example, let us have observations $(X_1,Y_1),\dots,(X_N,Y_N)$, where $Y_j$ are sampled from the normal distributions $N(a+bX_j,\ c+dX_j)$, respectively. Are there regression methods that allow us to estimate the parameters $a,b,c,d$?

  • $\begingroup$ You could look into gamma regression, but really you need to give us more details ... $\endgroup$ Commented Jan 20, 2021 at 2:26
  • 3
    $\begingroup$ Using gamma glms for variances is an established method, see stats.stackexchange.com/questions/500294/… $\endgroup$ Commented Jan 20, 2021 at 3:22
  • $\begingroup$ @kjetilbhalvorsen I have added an example of such a model. $\endgroup$
    – Viktor
    Commented Jan 20, 2021 at 4:19
  • 1
    $\begingroup$ In a recent question the variance was also a function of the regressors stats.stackexchange.com/a/505370 $\endgroup$ Commented Jan 22, 2021 at 18:55

1 Answer 1


For your example, for some parameter values $c,d$ the variance will be negative ... for that reason, in such models often is used a log link function for the variance. But such models (and many others) can be fitted with extensions of generalized linear models (glm's), also introducing link functions and linear predictors for the variance (and maybe even for other parameters.) One such family of models is known as gamlss see gamlss website for information.

For some examples Compare shape and scale parameters between Weibull distributions, Are there better approaches than the weighted mean?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.