When should a researcher choose logistic regression over linear regression?

Choosing logistic regression over linear regression is primarily driven by the nature of the outcome variable. Logistic regression is specifically designed to handle situations where the outcome variable is binary, meaning it has only two possible outcomes (for example, success/failure, yes/no, or 1/0). This method models the probability of one of the outcomes occurring as a function of the predictor variables.

The logistic function constrains the predicted probabilities to be between 0 and 1, which is essential for binary outcomes. In contrast, linear regression assumes that the outcome variable is continuous and can take any value along a spectrum, which is not suitable when the outcome is binary. In such cases, applying linear regression can lead to nonsensical predictions (such as probabilities less than 0 or greater than 1) and violate the assumptions underlying that model.

Using logistic regression allows researchers to interpret the estimated coefficients in terms of odds ratios, providing meaningful insights into the influence of predictor variables on the likelihood of the binary outcome. Therefore, when faced with a binary outcome variable, logistic regression is the appropriate choice for accurate modeling and interpretation of the data.