CUMULATIVE SEMANTIC INHIBITION
categories over and above the items that compose them. The interaction with Ordinal posi-
tion provides an estimate of the systematic variation in the amount of cumulative semantic
inhibition produced specifically by each category.
The comparison of the models’ estimates in Table 1 indicates that the magnitude and
significance of the main linear semantic effect is largely unaffected by the inclusion of this
new random factor. A formal comparison between models by means of a log-likelihood test
shows a significant improvement of the model’s fit, both from the inclusion of a variable
intercept across categories (χ2 (1) = 13.0, p < .001) and from a variable ordinal position
effect across categories (χ2 (1) = 16.5, p < .001). This indicates that, while it is true that
there is a main inhibitory effect of the ordinal position within the category, valid on average
for all categories, two additional factors not reported by Howard et al. (2006) need to be
considered. On the one hand, there is a significant random effect of category, meaning
that items in some categories are systematically faster than items in other categories. On
the other hand, the amount of inhibition provided by each occurrence of an item within a
category shows significant systematic variation across individual categories: Some categories
produce consistently more inhibition than others.
To clarify this finding, the main random effect of Categories and its interaction with
Ordinal position are plotted on the left panel of Figure 1. Three findings are noteworthy.
First, the overall speed for each category presents a considerable degree of variability, which
is estimated over and above item variability. Secondly, while all of the coefficients are
positive, there is also a significant variability in the magnitude of the linear cumulative effect
across categories. This indicates that every single category contributes to the cumulative
inhibition effect (with the possible exception of category 8, Body parts). Finally, the main
speed of a category is unrelated to its contribution to the semantic cumulative effect, with
no evidence for a correlation between the two estimates (r = .15, t(22) = .69, p = .49).
Discussion
The analysis we computed over the original dataset provides a confirmation of the
observations made by Howard et al. (2006) on the basis of data averaged by participants,
items and categories. They further show that the linear cumulative semantic inhibition
effect is present for all the categories that were tested(with one possible exception). In
other words, the cumulative semantic inhibition effect is in no way a consequence of the
trial position of successive members of a category, or a consequence of averaging across a
heterogeneous dataset. In addition, two new issues have arisen from our analyses. First,
there is a systematic variation in the overall speeds of items belonging to different categories.
Second, and more important, there is also a systematic variation in the strength of the
cumulative semantic inhibition across categories. Finally, the overall speed with which the
members of a category are named is unrelated to the strength of the cumulative inhibition
shown by that category. Therefore we cannot conclude that the variability in the size of the
cumulative effect is a mere consequence of variations in naming speed. One possible origin of
the variation in cumulative inhibition effect across categories may be the relationships among
the items that compose the categories. The theoretical interpretation of these findings will
be addressed in more detail in the General Discussion. For now, these observations set the
stage for a deeper exploration of the effect on a subset of the data.