What are the drawbacks of a linear model?

anshika-sharma-ac2988d4 · 28 August 2020 04:50

iftekar-patel-f1e6bf65 · 30 August 2020 01:38

Drawbacks (Assumptions) of linear model

Linear relationship.
Multivariate normality.
No or little multi collinearity.
No auto-correlation.
Homoscedasticity.
ref:https://medium.com/@kalyan77k/drawbacks-assumptions-of-linear-model-2e28d1c2f1d2

chirag-garg · 1 August 2021 06:15

The biggest limitation is the assumption of normality. The assumption statesQ that the sampling distribution is normally distributed. If this assumption is not met the results are not reliable, this (imho) is the largest source of error. Other problems are multicollinearity and heteroscedasticity. Big words meaning interaction between variables for the former and unequal variance within a single variable as the latter. Additionally it is possible to have nonlinear (squared and interaction) interactions. Lastly with everyone having access to regression programs it is easy to run a regression. Running the regression is only a step in the process. After a model is selected analysis of residuals is necessary but not always done.

These are all limitations to constructing a good regression model. If the modeler does all the proper analysis and validation of assumptions the result is a good predictive model.