The stepping action was divided into discrete phases: anticipatory, initial swing, terminal swing, and weight transfer.
17 Data were analyzed using a random effects population averaged model (Stata version 7.0; Stat Corp., College Station, TX). This multivariate model was obtained using the generalized least squares (GLS) random-effects estimator, which produces a matrix-weighted average of between-subjects and within-subjects results. An exchangeable correlation structure was judged to be appropriate, given the experimental design, and because of the exploratory nature of the study, no type 1 error adjustment of the alpha level was deemed necessary. Thus, level of significance was set at
P < 0.05. Factors of interest were incorporated sequentially, and their statistical significance was tested using a likelihood ratio test. Factors with a
P < 0.1 were provisionally retained, whereas those greater than 0.1 were dropped. The final model adopted was the most parsimonious one that was felt to adequately explain the data. A
t test revealed that young subjects were significantly taller than elderly subjects (
P = 0.01); therefore, subject height was deemed to be a confounding variable and was added as a confounder to the model. There was no difference in mass between the two groups (
P > 0.1).
P values quoted in the text of the paper are those associated with the specific terms in the final regression model, which were:
-
Vision: a fixed factor with 2 levels—normal (optimal correction) and blurred (diffuse blur);
-
Repetition: a fixed factor with 3 levels—trials one, two, and three;
-
Step height: a fixed factor with 3 levels—low, medium, and high step;
-
Phase: a fixed factor with four levels—anticipatory, initial swing, terminal swing, and weight transfer;
-
Age: a fixed factor with two levels—young and elderly;
-
Subject height: a confounding factor.