A question that has generated increasing interest in the past years is whether the default interletter spacing currently employed in word-processing software, books/e-books, or newspapers is optimal for reading (Gori & Facoetti, Reference Gori and Facoetti2015; McCandlis, Reference McCandliss2012; Montani, Facoetti, & Zorzi, Reference Montani, Facoetti and Zorzi2015; Perea & Gomez, Reference Perea and Gomez2012a, Reference Perea and Gomez2012b; Schneps, Thomson, Chen, Sonnert, Pomplun, Reference Schneps, Thomson, Chen, Sonnert and Pomplun2013; Schneps, Thomson, Sonnert et al., Reference Schneps, Thomson, Chen, Sonnert and Pomplun2013; Slattery & Rayner, Reference Slattery and Rayner2013; van den Boer & Hakvoort, Reference van den Boer and Hakvoort2015; Zorzi et al., Reference Zorzi, Barbiero, Facoetti, Lonciari, Carrozzi, Montico and Ziegler2012). We must keep in mind that font designers have not chosen the settings for the default interletter spacing based on empirical evidence from reading experiments.

As the recognition of objects (i.e., letters) is impaired by the presence of nearby objects (i.e., crowding effects; Bouma, Reference Bouma1970; see Moores, Cassim, & Talcott, Reference Moores, Cassim and Talcott2011), when letters in a word are too close (i.e., as in judge), the process of letter/word identification is hindered (i.e., Chung, Reference Chung2002; also see Gori & Facoetti, Reference Gori and Facoetti2015, for a review reading and crowding). Alternatively, when there is an exceedingly large increase in interletter spacing (i.e., as in judge), the words’ physical cohesion would break, thus hindering the process of word recognition (see Risko, Lanthier, & Besner, Reference Risko, Lanthier and Besner2011; Vinckier, Qiao, Pallier, Dehaene, & Cohen, Reference Vinckier, Qiao, Pallier, Dehaene and Cohen2011). But is the default interletter spacing optimal for reading? Data from a number of visual single-word recognition experiments have revealed that a small increase in interletter spacing, relative to the default settings, may help the process of word encoding. For example, in a lexical decision experiment (i.e., “is the stimulus a word or not?”) with adult skilled readers, Perea and Gomez (Reference Perea and Gomez2012a) manipulated parametrically the interletter spacing of words (–0.5, 0.0 [default], +0.5, +1.0, and +1.5 points [pt] to default; i.e., judge, judge, judge, judge, judge) using a 14-pt Times New Roman font. They found that word response times were faster for the conditions with extra spacing than for the conditions with condensed (–0.5) or default interletter spacing (0.0) (i.e., in terms of response times, judge > judge). To determine the locus of this phenomenon, Perea and Gomez (Reference Perea and Gomez2012a) conducted fits on a mathematical model, the diffusion model (see Ratcliff, Gomez, & McKoon, Reference Ratcliff, Gomez and McKoon2004). Perea and Gomez (Reference Perea and Gomez2012a) concluded that the facilitative effect of interletter spacing was due to letter/word encoding rather than to decisional processes.

While the findings obtained from laboratory word recognition tasks are important to unveil the process of lexical access, there is a caveat: readers rarely encounter isolated words; instead, they read words in the context of text forming sentences and paragraphs. Indeed, it is not straightforward to generalize the effects of interletter spacing found with isolated words to a normal reading scenario. While an increase in interletter spacing can help an early encoding stage—as deduced from the data from Perea and Gomez (Reference Perea and Gomez2012a); see also Perea, Moret-Tatay, and Gomez (Reference Perea, Moret-Tatay and Gomez2011), it may also have a deleterious effect during text reading. Visual acuity from the fixation point decreases very quickly, and hence, an increase in interletter spacing may produce a potentially detrimental effect: nearby words will be farther apart from fixation thus reducing the amount of parafoveal processing that is available during reading ^{Footnote 1}

Therefore, to answer the question of whether or not the default interletter spacing is optimal for reading, it is essential to consider the effects of interletter spacing during normal (silent) reading. The literature on this issue is scarce. In an unpublished study conducted by Tai, Sheedy, and Hayes (Reference Tai, Sheedy and Hayes2009), a group of adult skilled readers were asked to read a novel presented in various conditions ranging from of –1.75 points (compressed) to +2.0 points (expanded) with a 10-pt Verdana font. Tai et al. (Reference Tai, Sheedy and Hayes2009) found similar total reading times in all interletter spacing conditions. This null effect was the net result of two opposite effects that cancelled out each other: 1) the text with an increased interletter spacing received shorter fixations; and 2) the text with an increased interletter spacing received more fixations. Unfortunately, Tai et al. (Reference Tai, Sheedy and Hayes2009) did not indicate the precise details on the procedure they employed (i.e., the design of the experiment) and, furthermore, there were no indications that they measured reading comprehension. To carefully examine this issue, Perea and Gomez (Reference Perea and Gomez2012b) conducted an experiment to examine eye movement control during sentence reading while manipulating three interletter spacing conditions using the 14-pt Times New Roman font: sentences with the default interletter spacing (0.0), sentences with an expanded (+1 pt) interletter spacing, and sentences with an expanded (+1.5 pt) interletter spacing. Neither sentence reading times nor comprehension scores differed significantly for the three types of sentences. As occurred in the Tai et al. (Reference Tai, Sheedy and Hayes2009) experiment, fixation durations were shorter for the sentences with expanded interletter spacing than for the sentences with the default interletter spacing (228, 221, and 216 ms for the 0.0, +1.0, and +1.5 interletter spacing conditions, respectively). However, the number of fixations was higher in the sentences with a +1.5 interletter spacing than for the sentences with the default interletter spacing (9.9 vs. 9.5 fixations, respectively), and the net effect was a null effect of interletter spacing in total reading times. Likewise, Slattery and Rayner (Reference Slattery and Rayner2013) conducted a sentence reading experiment in which they manipulated four conditions of interletter spacing (–0.5, 0.0, +0.5 and +1.0) using Verdana and Times New Roman fonts. They found a quadratic trend of interletter spacing, in which the faster sentence reading times corresponded to the default spacing condition (1923, 1862, 1911, and 1909 ms, for the –0.5, 0.0, +0.5 and +1.0 conditions, respectively). As in the Perea and Gomez (Reference Perea and Gomez2012b) experiment, Slattery and Rayner (Reference Slattery and Rayner2013) found a linear trend in the average fixation durations (253, 247, 241, and 240 ms, for the –0.5, 0.0, +0.5 and +1.0 conditions, respectively) and more fixations in the sentences with extra interletter spacing (7.6, 7.6, 7.9, and 8.0, for the –0.5, 0.0, +0.5 and +1.0 conditions, respectively).

Other recent experiments have examined the role of interletter spacing when reading aloud a short story in developing readers (Perea, Panadero, Moret-Tatay, & Gomez, Reference Perea, Panadero, Moret-Tatay and Gomez2012; Zorzi et al., Reference Zorzi, Barbiero, Facoetti, Lonciari, Carrozzi, Montico and Ziegler2012)—note that the participants’ eye movements were not registered in either of these two studies. Perea et al. (Reference Perea, Panadero, Moret-Tatay and Gomez2012; Experiment 3) examined two interletter spacing conditions: default spacing (0.0) vs. increased spacing (+1.2) using the Times New Roman font. They found similar response times and comprehension rates in the two conditions for developing readers in Spanish—none of them had a reading disability diagnosis. Interestingly, an extra increase in interletter spacing did facilitate processing in children with developmental dyslexia (i.e., faster reading times and higher comprehension rates for the text with extra interletter spacing than with the default settings). Zorzi et al. (Reference Zorzi, Barbiero, Facoetti, Lonciari, Carrozzi, Montico and Ziegler2012) reported a similar pattern in French and Italian: they found an advantage of extra interletter spacing in reading times and comprehension rates when reading aloud short texts with dyslexic children, but not with two groups of normally developing readers—note that unlike the Perea et al. (Reference Perea, Panadero, Moret-Tatay and Gomez2012) experiment, Zorzi et al. (Reference Zorzi, Barbiero, Facoetti, Lonciari, Carrozzi, Montico and Ziegler2012) used two control groups (reading-level controls, and age-level controls). Thus, the data from Perea et al. (Reference Perea, Panadero, Moret-Tatay and Gomez2012) and Zorzi et al. (Reference Zorzi, Barbiero, Facoetti, Lonciari, Carrozzi, Montico and Ziegler2012) suggest that interletter spacing may be a relevant factor to add in future implementations of e-books (besides font, letter size, or background color) for readers with dyslexia—and possibly for readers with low vision (see McLeish, Reference McLeish2007).

The aim of the experiment is to examine the role of interletter spacing during normal silent reading with adult skilled readers in a more ecological scenario than in previous research. In the Perea and Gomez (Reference Perea and Gomez2012b) or Slattery and Rayner (Reference Slattery and Rayner2013) experiments, participants were asked to silently read isolated sentences. This is a very common procedure in eye movement research, as it may unveil when and where the eyes move—in particular when there are target words embedded in the sentences (i.e., high- vs. low-frequency words). However, one might argue that registering the eye movements from a set of unconnected sentences may not reflect the same processes as those occurring when one reads a text for comprehension. In the present experiment, participants were asked to read for comprehension two relatively short stories (367 words each) taken from a standardized test. One of the stories was presented with the default interletter spacing, whereas the other was presented with an expanded interletter spacing (+1.2 pt)—note that this latter interletter spacing condition was effective in previous research with isolated words and sentences (Perea et al., Reference Perea, Moret-Tatay and Gomez2011, Reference Perea, Panadero, Moret-Tatay and Gomez2012; Perea & Gomez, Reference Perea and Gomez2012a, Reference Perea and Gomez2012b) (see Table 1). All participants read the two stories of similar difficulty and length in a within-subject design. The participants’ eye movements were registered and the main dependent variables were the following: average fixation times, total reading times, and number of fixations. Comprehension scores were also examined.

Table 1.

Excerpt of one of the two texts with the two interletter spacing conditions

The predictions are clear. If the findings from sentence reading experiments (i.e., Perea & Gomez, Reference Perea and Gomez2012b; Slattery & Rayner, Reference Slattery and Rayner2013) can be generalized to text reading, one would expect shorter fixation durations when the text is presented with extra interletter spacing than when the text is presented with the default spacing; nonetheless, this effect may vanish when one considers the total reading times, as more fixations may be executed in the extra spacing condition. Alternatively, if eye movement guidance can readily adapt to an extra interletter settings during text reading while keeping an early word encoding advantage, one would expect not only shorter fixation durations in the text with extra interletter spacing than in the text with the default spacing, but also a parallel effect in the total reading times. This latter outcome would suggest that the default interletter spacing is not optimal for text reading.

Results

The eye movement data were analysed with EyeDoctor 0.6.5 software ^{Footnote 3} Extreme fixation duration times (beyond the 90–800 ms cutoffs) were excluded from the analyses (less than 1% of the data). We examined four dependent variables: average fixation time, total reading time, number of fixations, and comprehension scores. The participant’s means and standard errors in each dependent variable are presented in Table 2. In the statistical analyses, we included Group (list 1, list 2, list 3, list 4) as a dummy factor to remove the variance due to the counterbalanced lists—the outcome was virtually the same had this variable not been included in the analyses.

Table 2.

Measures for each of the conditions in the experiment: Total sentence reading time (in sec), mean fixation duration (in ms), number of fixations, and comprehension scores. Standard errors are presented between parentheses

Note: 95% CI refers to 95% confidence intervals.

We performed the statistical inference not only using the standard tests of null hypothesis testing (i.e., F-tests and p values), but we also computed Bayes Factors in JASP ^{Footnote 4} . Bayes factors offer an estimate of the support for a model (i.e., the model of the null hypothesis) relative to another model (i.e., the model of the alternate hypothesis) (Rouder, Speckman, Sun, Morey, & Iverson, 2012; see also Gomez & Perea, Reference Gomez and Perea2014). This allows us to obtain an estimate of the likelihood of the null vs. the alternate hypothesis. The expression BF₁₀ refers to the probability of the data given the alternate hypothesis (Model 1) relative to the probability of the data given the null hypothesis (Model 0): p(H₁|Data) / p(H₀|Data). For instance, BF₁₀ = 8 implies that the data are 8 times more likely under the alternate hypothesis than under the null hypothesis. Conversely, the expression BF₀₁ refers to the probability of the data given the null hypothesis (Model 0) relative to the probability of the data given the alternate hypothesis (Model 1). There are guidelines in the literature on how to interpret Bayes Factors (i.e., Jeffreys, Reference Jeffreys1961, p. 432): 0–3 would reflect “evidence [...], but not worth more than a bare mention”; 3–10 would reflect “substantial” evidence; 10–32 would reflect “strong” evidence, 32–100 would reflect “very strong” evidence, whereas values over 100 would reflect “decisive” evidence.

The average fixation duration was shorter for the text with extra interletter spacing than with the text with default spacing (237 vs. 245 ms, respectively), F(1, 12) = 8.38, MSE = 48.48, η²=. 41, p = .013, BF₁₀ = 4.18. The effect size was similar to that obtained in previous sentence reading experiments (i.e., Perea & Gomez, Reference Perea and Gomez2012b) and the Bayes Factor (BF₁₀ = 4.18) reflects reasonably substantial evidence in favor of a facilitative effect of interletter spacing (i.e., the alternate hypothesis). In addition, the number of fixations was only slightly higher in the text with extra interletter spacing than in the text with default spacing (375 vs. 371, respectively)—note that this effect did not approach significance (F < 1; BF₀₁ = 3.75). Finally, total reading times were only slightly shorter in the text with extra interletter spacing than in the text with default spacing (89.1 vs. 90.2 seconds for texts with extra vs. default interletter spacing, respectively, F < 1; BF₀₁ = 3.66). That is, the total reading data offer evidence for the null hypothesis (i.e., BF₀₁ > 3 in Jeffreys’, Reference Jeffreys1961 guidelines). We also inspected whether the lack of an effect of interletter spacing in total reading times was due to a small subset of participants showing the opposite effect. However, the percentage of participants who showed faster reading times in the extra spacing text than in the default spacing text (7 out of 16; 43.75%) was similar to the percentage of participants who showed faster reading times in the default setting than in the extra spacing text (9 out of 16, 56.25%) (sign test: p > .70).

The comprehension scores were 4.9 vs. 4.3 for the texts with extra vs. default interletter spacing; this difference did not approach significance, F(1, 12) = 1.61, p = .22.

Although the main focus of the analyses was on the total reading times and the fixation durations, we also collected other potentially informative eye movement variables. However, none of them approached statistical significance. For instance, the saccade length (both progressive and regressive) was similar in the texts with extra vs. default spacing (progressive saccades: 8.4 vs. 8.4, respectively; regressive saccades, 5.5 vs. 5.9, respectively; all ps > .25).