Table 4.

Bayesian binomial test for positional, forward, and backward bias in three-unit sequences (trigrams).

	Bigrams/trigrams	log(BF+0)	Posterior θ	95% CI
Positional bias, bigrams in trigrams	[HO + PH] + ♢	61.23	0.99	0.96–1
	[GR + PG] + ♢	47.32	0.92	0.87–0.96
	[PH + PB] + ♢	6.70	0.71	0.61–0.80
	♢ + [PH + PS]	−1.55	0.55	0.50–0.64
Forward bias, bigrams in trigrams	[GR + PG] → GR	5.93	0.68	0.59–0.77
	[HO_PH] → PS*	0.32	0.40	0.31–0.49
	[PH_PB] → PH*	1.43	0.36	0.25–0.47
	[PH_PS] → PB*	1.06	0.34	0.20–0.47
Backward bias, bigrams in trigrams	HO ← [PH_PS]	18.04	0.93	0.83–0.98
	HO ← [PH_PB]	3.88	0.78	0.61–0.91
	HO ← [GR_PG]*	−0.27	0.34	0.13–0.49

Results from the Bayesian binomial test evaluating positional, forward, and backward bias of bigrams in trigrams. The same parameters as described in Table 2 were used (see “Methods”). The number of successes over all trials for each bigram was defined according to the number of occurrences found in the most frequent position (head vs. tail) in the trigrams for the positional analysis (positional bias). The number of successes over all trials for each utterance was thus defined according to the number of times that υω was followed by ε (or that υω was preceded by ε), compared to the number of times that υω was followed by any other units (or that υω was preceded by any other unit) for the relationship analyses (forward and backward bias). Results are reported for the Bayesian factor log(BF₁₀), testing the hypothesis that the proportion of occurrences is higher than (unless specified) the default test value set at 0.5 (50%). Effect size estimates are reported as median posterior population (θ) with Credible Intervals (CI) set at 95%¹⁰². ^*Testing the null hypothesis that the proportion of occurrences in the most frequent position is lower than the default test value set at 0.5. ♢ = irrespective of unit type.