CUMULATIVE SEMANTIC INHIBITION
Table 1: Comparison of the fixed effects in linear mixed effect models of the log-transformed naming
latency in the full dataset (2568 observations from 24 participants naming 120 items in 24 categories).
H-model |
Fixed effect | ||
Ordinal position |
Lag |
Trial number | |
β |
3.81 ∙ 10-2 |
5.18 ∙ 10-4 |
2.26 ∙ 10-4 |
1 t(2564) |
7.22 |
.20 |
1.55 |
_______p |
< .001 |
.84 |
.12 |
β |
3.80 ∙ 10-2 |
6.59 ∙ 10-4 |
2.17 ∙ 10-4 |
2 t(2564) |
7.25 |
.26 |
.97 |
p |
< .001 |
.80 |
.33 |
β |
3.87 ∙ 10-2 |
7.16 ∙ 10-4 |
2.09 ∙ 10-4 |
3 t(2564) |
5.51 |
.28 |
.94 |
p |
< .001 |
.78 |
.35 |
Linear analysis of the dataset
We obtained the dataset used in the original study. This dataset comprised a total of
2568 trials (after the exclusion of filler trials, errors and responses identified as outliers). The
individual naming latencies were log-transformed to reduce skewness and approach a normal
distribution, then fitted to three linear mixed effect models with different combinations of
fixed, random, and mixed effects.
The first model, referred to as H-model 1, was intended to test the same hypothesis
than Howard et al. (2006), namely that there is a main linear effect of ordinal position in
the category independent of the lag between related trials and of the position of the trials
in the experiment. The fixed effects of interest were the Ordinal position in the category,
the Lag between the current trial and the previous trial from the same category, and the
Trial number (i.e. the ordinal position in the experiment). In addition, to be able to
handle trial-level data, participant and item identity were explicitly included as random
effects in the model. As can be seen on the first row of Table 1, Ordinal position has a
significant inhibitory effect, and no evidence is found for influences of Lag or Trial number.
In H-model 2 (second row of Table 1), the effect of Trial number was allowed to vary
between participants (see Howard et al., 2006). This was done by including an interaction
between the fixed effect of trial number and the random effect of participant (i.e., a mixed-
effect). A formal comparison of H-model 1 and 2 (namely, a log-likelihood test) shows a
significant improvement in the model’s fit (χ2(1) = 13.6, p = .001) while the estimates for
the theoretically relevant predictors remain largely unaffected. Together, these results are
fully concordant with those of Howard et al. (2006).
More interestingly, in the third model (H-model 3), we further estimated the cross-
categorical variability of the linear cumulative semantic inhibition effect. To do so, we
included Categories as an additional random effect, on top of participants and items. This
new random effect has 24 levels under which the items are nested. We computed its main
random effect, as well as its interaction with the inhibitory fixed effect of Ordinal position
(i.e., a mixed-effect). The main effect estimates possible systematic contributions of the