Figure 1
An illustration of the preview benefit (a) and phonological preview benefit (b) in English using the boundary technique (Rayner, 1975). The invisible boundary is illustrated with vertical dotted lines. In Chinese studies, the baseline for the phonological preview benefit effect is typically an unrelated-word preview (see the main text). In the classical preview benefit effect (panel a), different invalid preview masks can be used, e.g.: bear -> hurm (pseudoword), bear -> txvu (random letter mask), or bear -> tuna (unrelated word).
Figure 2
A flowchart of the literature search process (a) and results from the survey of the unpublished literature (b). The data for one study could not be used because it included participants who had taken part in another included experiment. “Other sources” refers to studies located through methods other than database searching (e.g., contact with authors).
Table 1
Alphabetical Studies Included in the Meta-analysis.
| Study | Language | N | Items/cond | Preview type | Stimuli | Effect size in ms (SE) | |
|---|---|---|---|---|---|---|---|
| FFD | GD | ||||||
| Vasilev et al. (2019), Exp.1† | English | 64 | 10 | PSH | Own | –5 (8.6) | 7 (11.7) |
| Vasilev et al. (2019), Exp.2† | English | 64 | 10 | PSH | Own | 3 (8.6) | 2 (11.6) |
| Barrington (2018)† | English | 23 | 16 | H | Chace 2005 | –3 (16.5) | –2 (21.7) |
| Blythe et al. (2018) | English | 23 | 8 | PSH | Own | 9 (10.5) | 13.5 (14.9) |
| Leinenger (2018), Exp.3 | English | 48 | 56 | H | Own | 6 (9.7) | 7 (12.4) |
| Leinenger (2018), Exp.4 | English | 48 | 60 | PSH | Own | 6 (9.1) | 8 (11.3) |
| Drieghe et al. (2016)† | English | 95 | N/A | H | Chace 2005 | –4 (3.3) | –3 (4.5) |
| Plummer (2015), Exp.1† | English | 68 | 14 | H | Own | N/A | 7 (7.15) |
| Plummer (2015), Exp.2a† | English | 51 | 18 | PSH | Own | N/A | 9.5 (6.3) |
| Plummer (2015), Exp.2b† | English | 53 | 24 | PSH | Own | N/A | 6.5 (6.7) |
| Tiffin-Richards et al. (2015) | German | 23 | 14 | PSH | Own | –4 (6.7) | –2 (8.6) |
| Worf (2015)† | English | 12 | 16 | H | Chace 2005 | 2 (14) | 5 (17.8) |
| Choi & Gordon (2014) | English | 48 | 12 | PSH | Pollatsek 1992/Lee 1999 | –7 (4.7) | –6 (6.9) |
| Bélanger et al. (2013) | English | 20 | 18 | H | Pollatsek 1992 | 10.5 (16) | 6 (20.7) |
| Winskel (2011) | Thai | 36 | 8 | H | Own | –5 (7.2) | –17 (12) |
| Chace et al. (2005) | English | 23 | 16 | H | Own | 7 (5.2) | 9 (7.6) |
| Murray & Flynn (2005)† | English | 30 | 7 | H | Own | –5 (15.2) | 22 (22.1) |
| Miellet & Sparrow (2004) | French | 13 | 20 | PSH | Own | 13 (15.5) | 15 (19.3) |
| Pollatsek et al. (1992), Exp.2 | English | 40 | 12 | H | Own | 20 (8) | 14 (11) |
[i] Note: Positive effects indicate evidence for phonological preview benefit. PSH: pseudo-homophone. H: homophone. SE: standard error of the mean difference. FFD: first fixation duration. GD: gaze duration. N: number of participants. Items/cond: number of items per condition. N/A: data not available.
† Unpublished study.
Table 2
Chinese Studies Included in the Meta-analysis.
| Study | N | Items/cond | Stimuli | Effect size in ms (SE) | |
|---|---|---|---|---|---|
| FFD | GD | ||||
| Luo et al. (2018), Exp.1 | 24 | 10 | Own | 12 (20.1) | 1 (39) |
| Luo et al. (2018), Exp.2 | 32 | 10 | Own | 13 (19.2) | 13 (30.3) |
| Pan et al. (2016), Exp.1 | 54 | 15 | Yan et al. (2009) | 12 (5.8) | 33 (8.6) |
| Pan et al. (2016), Exp.2 | 56 | 15 | Own | 5 (6.3) | 10 (8.6) |
| Tsai et al. (2012) | 50 | 10 | Yan et al. (2009)† | –2 (5.8) | –4 (9.1) |
| Yan et al. (2009) | 48 | 10 | Own | 3 (9.7) | 11 (19.7) |
| Tsai et al. (2004), Exp.1 | 20 | 24 | Own | 8 (7.4) | 22 (13.1) |
| Liu et al. (2002), Exp.1 | 27 | 10 | Own | N/A | 16 (15.9) |
[i] Note: Positive effects indicate evidence for phonological preview benefit. SE: standard error of the mean difference. FFD: first fixation duration. GD: gaze duration. N: number of participants. Items/cond: number of items per condition. N/A: data not available.
† Study stimuli were adapted for traditional Chinese readers.
Table 3
Posterior Effect Size Estimate of the Phonological Preview Benefit and 95% Credible Intervals from the Meta-Analysis Model.
| Type of analysis | k | Mean ES (ms) | 95% CrI | p(ES > 0|D) | p(ES > 3|D) | τ2 | I2 |
|---|---|---|---|---|---|---|---|
| FFD | |||||||
| Alphabetical | 16 | 1 | [–3.7, 6.3] | 0.64 | 0.20 | 22.4 | 5.0% |
| English-only | 13 | 1.9 | [–3.7, 8.3] | 0.73 | 0.33 | 33.2 | 15.6% |
| Chinese | 7 | 5.8 | [–2.2, 14.2] | 0.93 | 0.77 | 28.4 | 0.0% |
| GD | |||||||
| Alphabetical | 19 | 3.5 | [–1.2, 8.4] | 0.92 | 0.57 | 11.9 | 0.0% |
| English-only | 16 | 4.5 | [–0.4, 9.9] | 0.96 | 0.72 | 12.9 | 0.0% |
| Chinese | 8 | 14.1 | [–1.1, 29.1] | 0.96 | 0.93 | 173.9 | 26.2% |
[i] Note: FFD: first fixation duration. GD: gaze duration. k: number of studies included in the analysis. p(ES > 0|D): probability that the effect size is positive, given the data. p(ES > 3|D): probability that the effect is bigger than 3 ms, given the data. CrI: credible interval. τ2: estimated between-study variance. I2: percentage of variance that can be attributed to between-study heterogeneity.
Figure 3
Forest plot of the alphabetical meta-analysis model for FFD and GD. Plotted are the observed study means (and 95% CI) reported in the original papers and the posterior means (and 95% CrI) estimated by the Bayesian meta-analysis model. The size of squares is proportional to the weight of each study in the analysis (i.e., the inverse of the within-study variance of the sampling distribution).
Figure 4
Forest plot of the Chinese meta-analysis model for FFD and GD. Plotted are the observed study means (and 95% CI) reported in the original papers and the posterior means (and 95% CrI) estimated by the Bayesian meta-analysis model. The size of squares is proportional to the weight of each study in the analysis (i.e., the inverse of the within-study variance of the sampling distribution).
Figure 5
Statistical simulations with the estimated number of missing studies for the phonological PB effect in English (measured with gaze duration). The top row shows the three scenarios used to simulate the missing studies: a null effect (a), a small effect (b) and a larger effect (c). Each scenario represents a probability distribution from which the effect sizes of missing studies are assumed to come from. The second row shows the distribution of posterior effect sizes generated under each scenario with the estimated number of missing studies. The simulated effect sizes were generated by drawing a k number of missing studies from the respective distribution, adding them to the available studies from the paper, and repeating the meta-analysis model. All results are based on 10 000 Monte Carlo simulations.
Table 4
Results from 10 000 Monte Carlo Simulations of the Phonological PB Effect in Gaze Duration with Known and Estimated Number of Missing Studies (Standard Deviations in Parentheses).
| Simulation type | k | ES (available data) | Mean ES from 10 000 Monte Carlo simulations | ||
|---|---|---|---|---|---|
| Null effect | Small effect | Larger effect | |||
| Known missing | |||||
| Alphabetical | 4 | 3.5 | 2.97 (0.73) | 3.71 (0.76) | 4.46 (0.85) |
| English-only | 4 | 4.5 | 3.77 (0.81) | 4.59 (0.83) | 5.46 (0.94) |
| Chinese | 4 | 14.1 | 8.98 (1.67) | 10.81 (1.62) | 12.69 (1.59) |
| Estimated missing | |||||
| Alphabetical | 6 | 3.5 | 2.76 (0.85) | 3.78 (0.89) | 4.86 (0.98) |
| English-only | 5 | 4.5 | 3.65 (0.88) | 4.61 (0.91) | 5.62 (1) |
| Chinese | 2 | 14.1 | 11.06 (1.44) | 12.13 (1.4) | 13.29 (1.35) |
[i] Note: k: Number of simulated missing studies. ES: phonological PB effect size (in gaze durations).
Figure 6
Number of subjects and items needed to achieve different levels of statistical power to detect the phonological PB in gaze durations for English (panel a) and Chinese (panel b). Statistical power is based on LMMs with a “maximum” random effects structure (Barr et al., 2013). Warmer colours indicate more desirable levels of statistical power.
Figure B1
Funnel plots of the studies included in the meta-analysis. In the absence of bias, more precise studies are expected to appear narrowly at the top of the plot, whereas less precise studies are expected to scatter more widely at the bottom of the plot.