Introduction

Reading has long been used for therapeutic purposes; in particular, bibliotherapy has been applied to a range of mental health disorders, including depression, anxiety, mood and sleep disorders (Stip et al. 2020). In a study of stress levels among college students, Rizzoli et al. (2009) investigated the effects of yoga, humor, and reading on the stress levels of college students, as measured by the Daily Stress Inventory, systolic blood pressure, diastolic blood pressure, and heart rate. They found significant reductions in the physical signs of stress for all three interventions, but not for the self-reported stress levels. Levine et al. (2022) conducted a longitudinal study on university students, finding that recreational reading by students was associated with “reduced psychological distress” over the course of the academic year. Hunter and Gillen (2009) identified reading as one of a number of strategies used by elderly adults to cope with stress, while Finucane et al. (2021) found that reading a book in bed improved sleep. Thus, previous research indicates that reading can help alleviate stress and anxiety.

This study investigates the whether there is an association between reading and anxiety for the citizens of the Islands, an online virtual human population.1 Although the research literature suggests that reading would reduce anxiety levels, the research question was framed conservatively for this study to investigate an association in either direction. Study participants either read a book for 30 minutes or sat still for 30 minutes, and assessed their level of anxiety before and after the intervention. The parameter of interest is therefore \(\mu_{sit} - \mu_{read}\) where \(\mu_{sit}\) is the long-run population mean change in anxiety level after sitting still for thirty minutes, and where \(\mu_{read}\) is the long-run population mean change in anxiety level after reading a book for 30 minutes.

Data Collection Methods

The experimental units were adult Islanders from all three Islands. The populations of the three islands were sampled randomly so that the study will be generalizable to the Islands as a whole. To create these random samples, each city on each Island (six on Ironbard, nine on Providence, twelve on Bonne Santé) was put on a list for that Island and assigned a sequential number. The true random number generator at random.org was used to find four random numbers between one and the number of cities for the particular island.2 For each of the cities that were chosen, six house numbers were selected from the sampling frame of all houses in the city, by generating six random numbers between one and the number of houses in the city. For houses with more than one resident, the first householder listed was asked to participate in the study to ensure that all participants were over 18.

From the list of householders, each was assigned randomly to one intervention or the other (sit or read), so that causal inferences could be drawn from the results. This was done using the random sequence generator feature of random.org to generate two groups (by spreadsheet row number). Of the initial cohort of 72 households, one house was removed because it had no residents, and twenty householders declined to participate in the study, so the participation rate was 72%. To ensure that there would be enough participants, a second cohort was recruited by again randomly choosing four cities per island3 and four households per city. The first householder in each of these households was assigned randomly to one of the two groups. From the second cohort of 48 potential participants, 16 declined to participate, for a participation rate of 67%.

As data were being collected, it became clear that assigning groups randomly to the full list of potential participants was not the optimal way to assign groups; it would have been better to assign participants to a group randomly after they had consented to be in the study, because then we could be sure that the groups would be the same size. Fortunately, since the Islanders who declined to participate did so at random, the number of Islanders in each group was essentially equal. The final number of participants in the study was 83; there were 41 in the “sit” group and 42 in the “read” group.

Data were collected from participants in the course of about 35 minutes. After initial consent was granted, participants filled out a survey that asked how anxious, how restless, and how tense they were feeling, and also asked them to rate their anxiety on a scale from 1 to 10. Within a minute of completing this survey, participants either read a book for 30 minutes or sat still for 30 minutes, as randomly assigned. Within a minute or two of completing this activity, they took the initial survey a second time. (Using multiple browsers permitted the administration of the survey and task to a number of Islanders concurrently, with staggered start times.) For the purposes of this analysis, only the ratings of anxiety on the scale from 1 to 10 were used. The variable anxiety_change was computed by subtracting the pre-task anxiety score from the post-task anxiety score.

Descriptive Statistics

For this study, the explanatory variable is the binary categorical variable task, describing the task assigned to each group, sitting still or reading. The response variable is the quantitative variable anxiety_change, capturing the difference between the participants’ anxiety level after the task and the anxiety level before the task (computed by subtracting the pre-task score from the post-task score). The range of possible scores for the anxiety_change variable went from -9 to 9; negative value for anxiety_change indicates a reduction in anxiety after the task.

The table of summary statistics for the collected data shows that the mean value for anxiety_change for the read group is -2.464, but the mean value for anxiety_change for sit group is 0.317, and the median values are close to this as well at -2.5 (read) and 0 (sit). We see that on average the read group saw an decrease in anxiety levels, while the sit group saw on average a very slight increase. There is greater variability in anxiety_change for the sit group with a standard deviation of about 1.43 than for the read group with a standard deviation of about 1.05.

library(readr)
SitRead <- read_csv("/home/blauar/Documents/Projects/SitRead.csv")
favstats(anxiety_change ~ task, data = SitRead)
##   task min   Q1 median   Q3  max       mean       sd  n missing
## 1 read  -5 -3.0   -2.5 -1.5 -0.5 -2.4642857 1.049929 42       0
## 2  sit  -2 -0.5    0.0  1.5  3.0  0.3170732 1.426342 41       0
dotPlot(~anxiety_change | task, data = SitRead, cex = 0.75, width = .01, layout = c(1, 2), 
main="Comparison of dot plots")

The dotplots show that the distribution of anxiety_change values in the groups is quite different. For the read group, the distribution of anxiety_change values in the sample is more or less mound-shaped, possibly somewhat left skewed, while the distribution of anxiety change values in the sample for the sit group is less uniform; there is more variability in the sit group data, which corresponds with a larger standard deviation within that group.

bwplot(task ~ anxiety_change, horizontal = TRUE, data = SitRead, 
main="Comparison of box plots")

The boxplots of observed results suggest that while there may not be an association between sitting still for half an hour and reducing anxiety, there appears to be a clear association between reading for half an hour and reducing anxiety. The entire range of values for anxiety_change for the group that read a book is negative, with a minimum of -5 and a maximum of -0.5; the mean value is -2.464. This means that every study participant in the read group experienced some amount of anxiety reduction (a negative change in anxiety level). By contrast the range for the group that sat still has a minimum of -2 and a maximum of 3, with a mean of 0.317, meaning that on average, people in the group that sat still became slightly more anxious. Only the lower quartile of values for the sit group overlaps with the values of the read group.

Analysis of Results

For the population of adult Islanders, the parameter of interest is \(\mu_{sit} - \mu_{read}\) where \(\mu_{sit}\) is the long-run population mean change in anxiety level after sitting still for thirty minutes, and where \(\mu_{read}\) is the long-run population mean change in anxiety level after reading a book for 30 minutes. The null hypothesis is that there is no association between the type of task and the change in anxiety level. The alternative hypothesis is that there is an association between the type of task and the change in anxiety level.

For this setting, a type I error (rejecting the null hypothesis when it is true) would mean finding that there is a difference between the mean change in anxiety level after sitting and the mean change in anxiety level after reading even though there is no difference. With a type I error, we might recommend reading to reduce anxiety (based on the possible negative association) but that would not be effective. A type II error (failing to reject the null hypothesis when it is false) would mean finding that there is no difference between the mean change in anxiety level after sitting and the mean change in anxiety level after reading even though there is a difference. In this case, we might not recommend reading as a method to reduce anxiety, but that would be a missed opportunity.

Because participants were chosen randomly from the adult population of all of the Islands they can be considered a representative sample. As we analyze the study data, we will be able to apply the results to all adult Islanders.

From the summary statistics and boxplots, we saw that the mean change in anxiety was always negative for those who read a book, and that the value of the mean change in anxiety was always more negative for those who read a book than for those who sat still for 30 minutes. We can use R code to calculate the observed statistic.

diff(rev(mean(anxiety_change~task, data = SitRead)))
##      read 
## -2.781359

This observed statistic is the difference between the mean change in anxiety for the group that sat for 30 minutes and the group that read a book for 30 minutes: \(\bar{x}_{sit} - \bar{x}_{read} = -2.78\)

To determine whether this difference is unlikely to have happened by chance alone if the type of task is not associated with the change in anxiety level in the long run, we want to run theory-based statistical tests. In order to do so, we must ensure that the validity conditions for a two-sample t-test are met. These are that the quantitative variable should have a symmetric distribution in both groups OR we should have at least 20 observations in each group and the sample distributions should not be strongly skewed.

histogram(~ anxiety_change |task, data = SitRead, width = 0.5, layout = c(1, 2), 
main="Comparison of histograms")

We have over 40 observations in each group, but looking at histograms of our observed group distributions, we see they are not entirely symmetrical, although the medians and means for each group are close together as noted in the discussion of descriptive statistics. However, if we simulate a null distribution of the difference of the means, it is nicely symmetric and we can consider the validity condition met.

#simulated null distribution
set.rseed(498) 
SitRead.null <- do(1000) * diffmean(shuffle(anxiety_change) ~ task, data = SitRead)
dotPlot(~ diffmean, data = SitRead.null, width = 0.01, cex = .75,
groups = (diffmean <= -2.781|diffmean >= 2.781), 
main="Simulated null distribution of difference of means")

To determine whether the observed statistic would be likely to have occured by chance alone, we can compute a two-sided p-value using a theory-based approach.

pval(t.test(anxiety_change~task, data = SitRead))
##      p.value 
## 1.569746e-15

The p-value here is 1.57 x \(10^{-15}\), which means that there is almost zero probability of finding a value as extreme or more extreme than our observed statistic in the null distribution by random chance alone.

stat(t.test(anxiety_change ~ task, data = SitRead))
##         t 
## -10.09787

Our t-statistic is -10.1, which means that we would be likely to find a value as extreme or more extreme than our observed statistic more than 10 standard deviations in the left tail of the standardized null distribution.

The p-value and the t-statistic we computed both provide extremely strong evidence against the null hypothesis. We therefore make the statistical decision to reject the null hypothesis in favor of the alternative hypothesis: there is an association between the type of task and the change in anxiety level. We have statistically significant evidence that there is a genuine difference in mean anxiety change between those who read a book and those who sat still for 30 minutes. The negative value of the t-statistic indicates that the association between reading a book and anxiety levels is negative. Because our groups were assigned randomly, we can make a causal inference and conclude that reading a book lowers anxiety levels.

confint(t.test(anxiety_change~task, data = SitRead))
##   mean in group read mean in group sit     lower     upper level
## 1          -2.464286         0.3170732 -3.330253 -2.232465  0.95

To find a range of plausible values for the parameter \(\mu_{sit} - \mu_{read}\), we compute a confidence interval. Based on the theory-based calculations to compute a 95% confidence interval, we are 95% confident that the difference between the population-level mean change in anxiety level after sitting and the population-level mean change in anxiety level after reading is between -3.33 and -2.23. Zero is not included in our confidence interval. This implies that the null hypothesis is not plausible and is consistent with the statistical decision based on the p-value and the t-statistic: there is indeed an association between reading and anxiety level, and (since the observed statistic and t-statistic were negative) reading lowers anxiety levels. On a scale measuring anxiety from 1 to 10, lowering anxiety levels between 2.23 and 3.33 units is not only statistically significant, it is quite a large difference, suggesting that reading might be a quite effective way to reduce anxiety.

Conclusion

Inspired by works by Finucane et al. (2021), Hunter & Gillen (2009), Levine et al. (2022), and Rizzolo et al. (2009) who all examined the effects of reading on anxiety, stress levels, or the ability to sleep, this project investigated whether reading a book was associated with anxiety levels for adult Islanders. Sitting still was chosen as the control task; the study was designed to use a random sample from all three Islands and random assignment to the two groups so that results would be generalizable to all adult Islanders and cause and effect determinations could be made.

Data were collected from 83 Islanders (in two cohorts), and the value of the pre-task anxiety score was subtracted from the value of the post-task anxiety score to calculate the anxiety_change variable. The descriptive statistics from the collected data, especially the side by side boxplots, suggested that there was an association between reading and decreasing anxiety. The theory-based tests (calculations of the two-sided p-value, 1.57 x \(10^{-15}\) and t-statistic, -10.1) confirmed that there is statistically significant evidence against the null hypothesis (there is no association between the task and the change in anxiety level) and in favor of the alternative hypothesis (there is an association between the task and the change in anxiety level). The 95% confidence interval (-3.33, -2.23) supported the statistical decision to reject the null hypothesis in favor of the alternative hypothesis. On the basis of the confidence interval, we find that for adult Islanders, reading for half an hour can reduce anxiety levels by between 2.23 and 3.33 points on the 10-point scale for self-assessing anxiety.

The main lesson learned from data collection was that consent should have been obtained from potential participants before assigning experimental groups; this seems very obvious in retrospect, but before data collection began, it was unclear at what rate Islanders would decline to participate in the study. The overall design of the research question was also perhaps too cautious. Given the existence of studies pointing to the effectiveness of reading for reducing anxiety, the research question “does reading reduce anxiety levels?” would have been preferable even though the results showed that the association was negative.

There are many promising avenues for future research around the topic of reading and anxiety reduction. Future research could investigate the relative effectiveness of other kinds of tasks, such as exercise, listening to music, or playing an instrument, in reducing anxiety. Additional metrics such as blood pressure and heart rate could be introduced to measure the change in anxiety physiologically, as in Rizzolo et al. (2009). Most research has concentrated on the effectiveness of reading in reducing anxiety for one particular age group. Although Finucane et al. (2021) found that effects of reading before bed were not very different across different age groups, it might be interesting to investigate whether Islanders in different age groups respond differently to reading a book to reduce their anxiety levels. In this study there was no way to gauge what kind of book the Islanders were reading for 30 minutes, so a study on non-virtual human beings in the real world might consider whether the kind of reading material has an effect on anxiety reduction.

Works Cited

Finucane, E., O’Brien, A., Treweek, S., Newell, J., Das, K., Chapman, S., Wicks, P., Galvin, S., Healy, P., Biesty, L., Gillies, K., Noel-Storr, A., Gardner, H., O’Reilly, M. F., & Devane, D. (2021). Does reading a book in bed make a difference to sleep in comparison to not reading a book in bed? The People’s Trial—an online, pragmatic, randomised trial. Trials, 22(1), 873. https://doi.org/10.1186/s13063-021-05831-3

Hunter, I. R., & Gillen, M. C. (2009). Stress Coping Mechanisms in Elderly Adults: An Initial Study of Recreational and Other Coping Behaviors in Nursing Home Patients. Adultspan Journal, 8(1), 43–53. https://doi.org/10.1002/j.2161-0029.2009.tb00056.x

Levine, S. L., Cherrier, S., Holding, A. C., & Koestner, R. (2022). For the love of reading: Recreational reading reduces psychological distress in college students and autonomous motivation is the key. Journal of American College Health, 70(1), 158–164. https://doi.org/10.1080/07448481.2020.1728280

Rizzolo, D., Zipp, G. P., Stiskal, D., & Simpkins, S. (2009). Stress Management Strategies For Students: The Immediate Effects Of Yoga, Humor, And Reading On Stress. Journal of College Teaching and Learning, 6(8), 79–88. https://www.proquest.com/docview/218893605/abstract/607D4AB2BBE045D0PQ/1

Stip, E., Östlundh, L., & Abdel Aziz, K. (2020). Bibliotherapy: Reading OVID During COVID. Frontiers in Psychiatry, 1–8. https://www.frontiersin.org/articles/10.3389/fpsyt.2020.567539

  1. http://islands.smp.uq.edu.au/↩︎

  2. Ironbard: Helluland, Helvig, Blonduos, Bjurholm; Providence: Takazaki, Nelson, Reading, Hayarano; Bonne Santé: Mahuti, Riroua, Eden, Maeva↩︎

  3. Ironbard: Vardo, Bjurholm, Hofn, Blonduos; Providence: Nelson, Hayarano, Arcadia, Akkeshi; Bonne Santé: Talu, Vaiku, Kinsale, Nidoma. Note that in this second cohort, Bjurholm, Blonduos, Nelson, and Hayarano were represented a second time.↩︎