2.6. Data Analysis

Data analysis was performed using IBM SPSS Statistics 23 (IBM Corporation, Armonk, NY, USA). First, we screened the data on test-meal energy intake for outliers. We calculated the total test-meal energy intake (cheese scones plus chocolate-chip muffins) for the water preload as the nominal control condition, minus the total energy intake for each of the other preloads and converted these data to z-scores. There were no z-scores lying outside of our criteria of z > 3.29 or z < −3.29 (i.e., no scores falling outside 99.9% of a normal distribution), and therefore, no participants were excluded from the data analyses.

As described elsewhere [22], we calculated individual energy intake compensation scores by subtracting test-meal energy intake after the energy-containing preload from the test-meal energy intake after the water preload, dividing by 569 kJ (the difference in the energy value of the energy-containing preload and the water preload) and multiplying by 100. We calculated change in fullness, hunger and thirst ratings by subtracting the ratings made before consumption of the preload (baseline) from the ratings made immediately before starting the test-meal. We calculated the ratio of sweet food to total food consumed in the test-meal (kJ chocolate-chip muffins consumed divided by kJ chocolate-chip muffins plus kJ cheese scones consumed).

We analysed the data on the characteristics of the preloads (liking, etc.) and time taken to consume the preloads using single factor (preload, 5 levels), repeated measures ANOVAs. Differences between all pairs of means were tested using the least significant difference (LSD) test. We analysed the energy-intake compensation scores using a two-way mixed factor analysis of variance (ANOVA) with preload (4 levels) being the within subject factor and preload to test-meal interval (2 levels) being the between subjects factor. The data on changes in fullness, hunger and thirst were analysed using two-way mixed factor ANOVAs, with preload (5 levels) being the within subject factor and preload to test-meal interval (2 levels) being the between subjects factor. We followed the two-factor analyses with single factor, repeated measures ANOVAs, conducted separately for the 2-min and 2-h interval groups, and then tested for differences between all pairs of means using the LSD test. Where appropriate, the Greehouse–Geisser correction was applied for effects involving preload, with corrected p-values reported. We recognise that the LSD test does not correct for multiple comparisons (family-wise error rate), but it has a high level of power to detect differences [23]. Accordingly, we interpreted the results of the LSD tests conservatively; that is, we describe and interpret patterns of results across the preloads (ordered from water to fruit salad) rather than contrasting all pairs of preloads.