Keywords: two-way tables, additive model, least-absolute-value estimation



Least-absolute-value Estimation in Two-way Tables and its Application to Explaining Variation in Flight Times



Natasha Yakovchuk

Ph.D. student

Department of Decision Sciences and Engineering Systems

Rensselaer Polytechnic Institute


Thomas R. Willemain, Ph.D.

Professor

Department of Decision Sciences and Engineering Systems

Rensselaer Polytechnic Institute




Abstract



Data structures known as two-way tables or two-way layouts have each of two factors varying separately, and a single response value observed for each combination of these factors. Such tables arise in many fields like socioeconomic studies, engineering experiments or transportation problems. The easiest and the most natural way to interpret the joint contribution of the factors is by decomposing the response in each cell as

response = common effect + row effect + column effect + residual


The most used technique for estimating effects is classical least squares, as in ANOVA.

We consider a two-way table of temporal deviations from flight plans for commercial US flights. We interpret the factors as system, origin and destination effects, and the residuals as en route effects. Classical least-squares estimation fails to give good estimates in this case because it is not resistant to outliers and 'holes' in the table. We illustrate the use of more robust least-absolute-values (LAV) estimation with the aviation data and present graphical displays derived from the LAV estimates.