A New Method for Analyzing Time Intensity Curves
This presentation introduces a novel statistical method for analyzing Time Intensity (TI) curves using an EM (expectation-maximization) algorithm combined with smoothing splines. The approach addresses a core constraint of TI data: curves monotonically increase to Tmax then monotonically decrease. The researchers model observed TI values as the maximum (ascending phase) or minimum (descending phase) of adjacent latent values, using truncated normal distributions and Markovian error terms. Two model variants were developed—homoscedastic (constant variance across time) and heteroscedastic (time-varying variance). Simulations validated the method, and real fruit liqueur data demonstrated practical applications including panelist clustering based on TI response patterns. The homoscedastic model was recommended for obtaining smooth, interpretable curves.
Browne, R. P., McNicholas, P. D. Castura, J. C., & Li, M. (2012). A New Method for Analyzing Time Intensity Curves. In: 11th Meeting of the Sensometric Society. 11-13 July. Rennes, France.