Application of OLS Regression for the Prediction of Labour Time in Manufacturing. A Comparison of Subset Selection and Exhaustive Search for Model Building

Ordinary least squares (OLS) regression is widely used to support cost estimating decisions. OLS regression uses past production data in order to estimate production costs (with both interval and point predictions) for similar new products, without requiring detailed information. In a regression model, the dependent variable is the production cost (e.g., the quantity of labour time) and the independent variables are the product parameters. Usually, these variables consist in some features of the product (performances, morphological characteristics, type of materials used), which are supposed to influence mainly the final production cost. OLS regression model is built through the application of a procedure in order to select variables that are of significant effect on the response. The present paper aims at illustrating the compared results of the application of two approaches for model selection in OLS regression – respectively, subset selection based on stepwise approach and exhaustive search based on prediction R-square coefficient – for the estimation of labour time in manufacturing. A real case study is used as reference within the paper. In particular, labour time data concerned the development of new sofa models, as performed in a manufacturing firm working in Southern Italy, were exploited. The results of this study give insight into the efficacy of OLS regression for labour time estimation during a new product cost development process, and evidence on the effect that the procedure, which is used to select variables, may have on the prediction ability of the OLS equation.