Motivation
In the previous subsection, I have indicated that in the current experiment faster readers experienced more mindless reading episodes. One plausible explanations of this result is that if reading speed is a proxy for reading skill, then reading in general should be easier for faster readers. This is important, because the incidence of mind-wandering has been shown to decrease in the context of cognitively demanding tasks (Smallwood & Schooler,
2006) and if that is also the case with reading then faster readers should be expected to mind-wander more.
However, because this result is correlational in nature, an alternative explanation ex- ists. Namely, it could be that subjects who experienced more mindless reading episodes read at higher average speeds because of that. This would be consistent with the hy- pothesis that text processing is shallower during mindless reading as suggested earlier by Reichle, Reineberg, & Schooler (2010) and Schad, Nuthmann, & Engbert (2012) and as I show in this dissertation (Sections 4.6 and 4.10). More specifically, that hypothesis is based on the observation that the effects of lexical and linguistic word characteristics are attenuated when a mind of a reader has wandered. That is, as eye movements are programmed, the where decision (which depends largely on low-level visual factors) re- mains in effect while the when decision (which depends largely on lexical and linguistic properties of words) intervenes to a lesser degree or not at all (for a review of the when- and-where, see Rayner, 1998). Importantly, Rayner & McConkie (1976) has shown that these two decisions can in fact be made somewhat independently.
Finally, it is possible that the combination of the two explanations I have offered above determines the interaction between reading speed and propensity to mind-wander. In this section, however, I investigate the second of the two explanations only. I do that by checking if reading speeds during normal and mindless reading differ. If mindless reading proceeds faster than normal reading then that would support the hypothesis that text processing during mindless reading is impoverished.
Models and Data
To check if the mean reading speed was different in normal and mindless reading, I fitted a series of 13 identical models (one per time window), each given as
read.spd.rij|si ∼ Normal(µij, σij2), µij = β0+ β1(rij) + si, Var[read.spd.rij] = σij2, si ∼ iid Normal(0, σ2s), i ∈ {1, 2, . . . , 77}, j ∈ {1, 2, 3}, (4.5)
where µ is the mean being modeled, s is the subject random effect, i is the subject index, j is the reading mode (i.e., N, P, and S) index.
Results
Figure 13 shows the results of fitting model (4.5). The first row of the figure shows the mean reading speeds in normal reading (N), probe-caught mindless reading (P), and self- caught mindless reading (S). The second row shows the differences between these three means and the mean for normal reading; that makes it easy to see how mindless reading diverges from attentive reading. The three subsequent rows show p-values associated with the three comparisons between then N, P, and S conditions (i.e., N-P, N-S, and P-S). Finally, the two last rows report model diagnostic information in the form of the variance of Pearson residuals (ρ2
pr; values close to one indicate no dispersion problems) and the
root mean square error (RMSE).
Because I use this figure template in several other analyses I discuss in the remainder of this dissertation I encourage the reader to understand it before moving on. I digress further to elucidate how I interpret this and other figures which report p-values for a number of time windows. Instead of cherry-picking statistically significant differences I require significant differences to be stable across several adjacent time windows. Ad- ditionally, I exercise even greater caution in interpreting results from the narrowest time windows because they are based on the smallest amount of data.
As is evident from the figure (rows 2–5), when reading mindfully, readers proceeded at 7.5–12.5 words per minute slower compared to the two mindless reading conditions. However, while this was the case for both probe-caught and self-caught mindless reading, only the self-caught mindless reading difference was statistically reliable and only in time windows of 15–40s. Because the number of probe-caught mindless reading episodes was substantially smaller (roughly half the amount) than self-caught mindless reading, this lack of statistical significance involving probe-caught mindless reading could be a result of insufficient statistical power, but there is no way of being sure.
The first row of the figure looks rather suspicious because it appears that reading speed is dropping proportionally to the time window size. However, this is not what the figure actually shows. Instead, this drop is associated with the intrinsic difficulty in calculating reading speed in the context of natural reading of ecologically valid text. The formula I used to calculate that speed was
read.spd.r = number.of.words.seen time.window.size · 60.
Clearly, the more words a reader sees and the narrower the time window the higher the reading speed. However, re-reading becomes more likely with the increase in time win- dow size. That is, while the denominator of this formula increases proportionally to the time window size, the nominator does not increase nearly as fast because re-reading in- volves words which have already been seen (i.e., re-reading a word does not result in a larger nominator). That is why reading speed appears to drop in the figure. How- ever, because of the relative stability of (1) the differences between normal reading and both kinds of mindless reading and (2) the differences between the two kinds of mind- less reading themselves, it seems reasonable to assume that the formula I used captures a stable reading speed as modulated by re-reading.
Conclusions and Discussion
I have found reading speed to be smaller during normal reading as compared to self- caught mindless reading in time windows of 15–40s. The difference between the two speeds was about 10 words per minute. The fact that readers sped up while being mind-
less could plausibly be a result of the more superficial text processing, one in which the effects of perceptual word variables (e.g., word length) are used to program eye move- ments while the effects of lexical word variables (e.g., word frequency) are attenuated.
Figure 13: Reading speed (read.spd; in words per minute) for normal reading (N), probe-caught mindless reading (P), and self-caught mindless reading (S). Based on fit- ting model (4.5).