Under what research conditions would you choose multiple regression over simple linear regression? Explain the assumptions of multiple regression with the help of examples.
15 marks · directive: Explain
The core of a good answer
Multiple regression is chosen when a criterion depends on two or more predictors, when we want each predictor's unique contribution while controlling for the others, or when we need better prediction than one variable gives; it assumes a linear relationship, independent errors, homoscedasticity, normally distributed residuals, no severe multicollinearity, no undue outliers and reliably measured predictors.
An outline written by upsc.help, not an official UPSC answer; every fact checked by a separate review. How we work
What the question asks
The question has two parts: conditions for choosing multiple over simple regression, and the assumptions with examples. Keep a running psychological example and show how each assumption is checked.
Answer outline
Introduction
Simple linear regression predicts a criterion Y from one predictor X (Y = a + bX). Multiple regression extends this to several predictors: Y = b0 + b1X1 + b2X2 + ... + bkXk + e.
When to choose multiple regression
- The outcome is multiply determined: academic achievement depends on intelligence, study hours, motivation and socio-economic status.
- To estimate the unique effect of a predictor while holding others constant (statistical control of confounds).
- To compare predictors on a common scale by standardising them.
Assumptions with examples
- Linearity: each predictor relates linearly to the criterion; a predictor such as anxiety may not relate to performance in a straight line.
- Independence of errors: residuals of different participants are unrelated; the model assumes independent, identically distributed errors.
- Homoscedasticity: the errors have the same variance throughout; tested on the residuals with, for example, the Breusch-Pagan test.
- Normality of residuals: the classical model assumes normally distributed errors.
- No severe multicollinearity: predictors such as verbal IQ and vocabulary score that correlate very highly make coefficients unstable; checked with tolerance or the variance inflation factor.
- No undue outliers: extreme cases can distort least squares estimates, and methods robust to outliers can be used.
- Predictors measured without much error: error in the predictors biases the coefficient estimates.
Conclusion
Multiple regression is the natural tool for the multi-causal questions of psychology, but its results are meaningful only when its assumptions are checked and reported.
Examples and data to cite
- Predicting job performance from a cognitive ability test and a conscientiousness scale together.
- Predicting depression scores from stress, social support and coping style.
Facts to remember
- A regression coefficient is the expected change in Y for a one-unit change in that predictor when the other predictors are held fixed.
- The variance inflation factor measures how much the variance of a coefficient is increased by collinearity; tolerance is its reciprocal.