The ongoing debate surrounding the efficacy of fidget toys in academic settings remains a topic of considerable interest. Advocates, exemplified by companies such as Legacy Toys, contend that these products alleviate stress and anxiety while enhancing focus and engagement (Legacy Toys)1. Numerous studies have looked into the effects of fidget toys within academic environments, with particular emphasis on their impact on attention, academic performance, and anxiety levels, especially among students with ADHD. For instance, a study examining the efficacy of fidget toys in a School Setting for Children with Attention Difficulties and Hyperactivity found that, among third graders, these toys may heighten attention and on-task behavior for select students based on individual characteristics (Rohrberger, 2011)2. Conversely, research indicates that, concerning specific academic domains like mathematics, fidget toys such as cubes fail to enhance performance and study outcomes for students with attention difficulties (Corley et al., 2023)3. Nonetheless, conflicting findings persist, with some studies reporting no discernible effects of fidget toys on attention and anxiety levels (Kirby,2011)4. This diversity of outcomes highlights a notable lack of consensus. However, it is essential to acknowledge the limitations inherent in these investigations, notably their predominant focus on lower-level education, relatively small sample sizes, and the need for further exploration of various academic domains related to memory. Given the increased interest in fidget toys and the gaps related to previous studies, this study aims to explore their effects on memory levels of individuals 18 years of age or older. Specifically, the research question addressed is: “How do different types of fidget toys, specifically poppit and spinner, affect individuals’ memory, as assessed by their error scores?”
Primary Hypothesis: It is hypothesized that individuals will have different number or errors depending on which type of fidget toy they watched while doing the memory test.
Secondary Hypotheses: It is also hypothesized that individuals will have significantly different number of error depending on the type of memory test being taken (image/visual, words, numbers).
This experimental study, conducted online via the Qualtrics survey platform, employed a randomized split-plot design to investigate the impact of different fidget toys on individuals’ memory performance. The study focused on two key factors: the type of fidget toy and the type of memory task.
The between-subjects factor under investigation was the type of fidget toy, with two levels: poppit and fidget spinner. Meanwhile, the within-subjects factor explored was memory type, which encompassed three levels: image, numbers, and words.
Participants were asked beforehand to rate their memory proficiency for images, words, and numbers on a Likert scale from “Not well at all” to “Extremely well.”
Participants’ errors on the memory test, scored out of seven, served as the response variable, providing a quantitative measure of memory performance across the different conditions. Each participant represented a “block” in the study design, ensuring control for individual differences.
The experimental units primarily comprised of students enrolled in Smith College’s SDS 291 Research, Design, & Analysis course, and those outside the course who were required to be at least 18 years old. In total, 38 participants or experimental units were included in the studies analysis after data cleaning, offering a representative sample for analysis.
Materials/Stimuli
The stimulus in this study comprised of GIFs of a poppit and a fidget spinner, images of donuts, a list of words, and numbers, representing the different memory tests:
GIFs: Participants were presented with GIFs of a poppit or a fidget spinner thoughout their memory test to represent the two types of fidget toys in the study.
Image Test: All Participants were presented with three images of donuts to remember and had to recall them within a total of seven donuts.
Number Test: All Participants were presented with five 5-digit numbers to remember and had to recall them within a total of seven numbers.
Word Test: All Participants were presented with five words to remember and had to recall them within a total of seven words.
Procedure and Data Collection
Participants first had to read and agree to a consent form to participate in this study. After agreeing, participants were asked to rate their memory proficiency for images, words, and numbers on a Likert scale from “Not well at all” to “Extremely well”. Then participants were randomly assigned to one of the two fidget toy levels using Qualtrics’ randomization feature. Participants performed three memory tests while their representative fidget GIF was playing in the background. Each memory test corresponded to a different memory type (image, numbers, and words) each representing different learning/academic domains. The order in which each memory type test was presented to the participant was randomized as well. For each test, participants were presented with stimuli (images of donuts, lists of numbers, and words) and asked to recall them within a total of 7 items. Their number of errors were scored later for examination. The data collection occurred in the SDS 291 classroom and online. It is noteworthy that participant feedback was collected anonymously to maintain confidentiality and foster honest responses.
This section presents the results of the analysis conducted to examine the impact of different fidget toys on individuals’ memory performance. ANOVA was primarily used to compare the mean error scores across the three memory tests: image, numbers, and words, and to compare the mean error scores across the two types of fidget toys: poppit and spinner. Interaction effect between fidget toy and memory was also examined.
The boxplot (Figure 1) provides a visual representation of the distribution of error scores across the three memory types, grouped by fidget toy type. Here’s a summary of the observations:
Image Memory Test: The boxplot shows that most participants had 1 error in the image memory test, with a few outliers indicating higher and or no error rates. Overall, the spread of error scores appears relatively tight, suggesting consistency in performance across participants.
Number Memory Test: The boxplot illustrates the highest variability in error scores among the three memory tests. While the median error rate is lower than that of the word memory test, it still appears higher than the image memory test.
Word Memory Test: The boxplot reveals the highest median error rate compared to the image and number memory tests. While the spread of error scores seems similar to that of the number memory test, the median error rate is noticeably higher.
Regarding the influence of fidget toy type (poppit and spinner) on error rates, the boxplot suggests consistency across all memory tests. There are no apparent differences in error rates between participants using poppit and spinner fidget toys within each memory test. This also suggest that there is no interaction between fidget type and memory type.
## Boxplot (Normality is violated)
pivot_longer_df |>
ggplot(aes(type, error, color = fidget_type))+
geom_boxplot() +
labs(x = "Memory Test", y ="Number of Errors", color = "Fidget Type", title = "Boxplot of Memory Test, Fidget Type, and Number of Errors")
#The assumption of normality is clearly violated (skewness, especially for words)
Summary statistics for both memory type and fidget type were also calculated and presented in Table 1 and Table 2. 18 participants were assigned the poppit while 20 participants were assigned the spinner.
ds_memory<- pivot_longer_df %>%
group_by(type) %>%
summarise(n = n(),
memory_mean = mean(error),
memory_sd = sd(error))
kable(ds_memory)| type | n | memory_mean | memory_sd |
|---|---|---|---|
| image | 38 | 0.9473684 | 0.6554292 |
| number | 38 | 4.8157895 | 1.0617493 |
| words | 38 | 5.8157895 | 0.5122955 |
ds_fidget <- pivot_longer_df %>%
group_by(fidget_type) %>%
summarise(n= n()/3,
fidget_mean = mean(error),
fidget_sd = sd(error))
kable(ds_fidget)| fidget_type | n | fidget_mean | fidget_sd |
|---|---|---|---|
| POPPIT | 18 | 3.907407 | 2.225652 |
| SPINNER | 20 | 3.816667 | 2.281032 |
When analyzing memory types, it’s clear that there are notable differences in means and standard deviations among groups. Specifically, the error score mean is highest for the Words group (5.81), while the Image group exhibits the lowest mean error score (0.94). Moreover, the Number group shows the highest variability (sd = 1.06), exceeding both Image and Words groups. These findings align with previous observations made regarding the distribution of scores as depicted in the boxplots.
For fidget types, a strikingly different pattern is shown with consistent means and standard deviations across groups, regardless of whether participants used a poppit or a spinner. Furthermore, the maximum standard deviation is two times less than the lowest standard deviation (1.02). These observations indicate consistent variability across all groups, reflecting a stable and uniform distribution of scores among participants, regardless of their fidget type.
From the boxplots in Figure 1 and the summary statistics in Table 1 and Table 2, it’s clear that Normality Assumption and the Standard Deviation Assumption are potentially violated. The presence of outliers and asymmetrical boxplots, particularly in image and words distributions, suggests a deviation from normality. Additionally, for memory type there is a difference in standard deviations, with the maximum being more than twice the smallest (2.07), indicates inconsistency across groups.This suggests a need, for some kind of transformation.
Since both Normality and Standard Deviation Assumptions were violated, a diagnostic transformation was performed in hopes of improving the properties of the data for analysis.The transformed boxplots can be seen in Figure 2, and the transformed summary statistics are provided in Table 3.
#Diagnostic Transformation
p1<- qplot(x = log(memory_mean), y = log(memory_sd), data = ds_memory) +
geom_smooth(method = lm, se = 0) +
labs(title = "Diagnostic Transformation")
transf <- lm(log(memory_sd) ~ log(memory_mean), data = ds_memory)
p <- 1-(0.03344)
# Replot using Transformation
p2<-qplot(x = type, y = error^p, data = pivot_longer_df, color = fidget_type, geom = "boxplot") +
labs(title = "Diagnostic Transformed Boxplot",
y = "Number or Errors^p",
x = "Memory Type")
p1+p2
#Retest S Condition
trans_ds <- pivot_longer_df |>
group_by(type)|>
summarise(n = n(), # each have 12 observations
mean = mean(error)^p,
sd = sd(error^p))
kable(trans_ds)| type | n | mean | sd |
|---|---|---|---|
| image | 38 | 0.9490828 | 0.6353014 |
| number | 38 | 4.5691887 | 0.9749560 |
| words | 38 | 5.4832767 | 0.4671123 |
After applying diagnostic transformations, we found that it didn’t notably enhance the normality conditions or rectify the violation of the Standard Deviation Assumption. Notably, the standard deviation remains inconsistent across groups, with the maximum exceeding twice the value of the lowest standard deviation (2.08). This suggests we run our ANOVA using our original variables of memory type and fidget type.
mod <- lmer(error ~ type*fidget_type + (1|response_id), data = pivot_longer_df)
anova(mod)?(caption)
Type III Analysis of Variance Table with Satterthwaite's method
Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
type 500.05 250.023 2 108 403.6589 <2e-16 ***
fidget_type 0.23 0.234 1 108 0.3778 0.5401
type:fidget_type 0.19 0.094 2 108 0.1512 0.8598
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1The results from our ANOVA can be seen in ?@tbl-4. According to our ANOVA results, there is evidence to suggest that the factor memory type has an effect on our response variable errors F(2, 402.115), p = <2e-16. Additionally, we found no evidence that fidget type has an effect on errors F(1, 0.368), p = 0.545 and neither does the interaction between memory type and fidget type F(2, 0.150), p = 0.861. It is important to note that our Standard Deviation, Normality, and Additive conditions for this model was violated thus our ANOVA interpretation should be taken with some considerations.
kable(confint(mod))| 2.5 % | 97.5 % | |
|---|---|---|
| .sig01 | 0.0000000 | 0.3450227 |
| .sigma | 0.6764064 | 0.8773301 |
| (Intercept) | 0.6986749 | 1.4124362 |
| typenumber | 3.2730723 | 4.2824833 |
| typewords | 4.2730723 | 5.2824833 |
| fidget_typeSPINNER | -0.6974816 | 0.2863705 |
| typenumber:fidget_typeSPINNER | -0.5234663 | 0.8679107 |
| typewords:fidget_typeSPINNER | -0.5234663 | 0.8679107 |
The estimates 95% confidence intervals can be seen in Table 4. If a confidence interval does not contain 0 for a coefficient estimate, it suggests that there is a statistically significant effect associated with that variable at the 95% level, and therefore reject the null.
Significant coefficients include intercept (baseline for typeimage), typenumber, and typewords.
Intercept (image type) confidence interval suggests that the true baseline error rate could range from 0.699 to 1.412.
Number type coefficient represents the effect of the number memory test compared to the reference group (image). The confidence interval suggests that the true difference in error rates between the number and image memory tests could range from 3.273 to 4.282.
Similarly, the word type coefficient represents the effect of the word memory test compared to the reference group. The confidence interval suggests that the true difference in error rates between the word and image memory tests could range from 4.273 to 5.282.
model1 <- aov(error ~ type * fidget_type, data = pivot_longer_df)
#perform Fisher's LSD
kable((LSD.test(model1,"type")$groups))| error | groups | |
|---|---|---|
| words | 5.8157895 | a |
| number | 4.8157895 | b |
| image | 0.9473684 | c |
# Compute least squares means
lsmeans_obj <- lsmeans(mod, ~ type)
# Perform pairwise comparison test
lsd_results <- pairs(lsmeans_obj)
kable(lsd_results)| contrast | estimate | SE | df | t.ratio | p.value |
|---|---|---|---|---|---|
| image - number | -3.863889 | 0.1808043 | 72 | -21.370561 | 0.0e+00 |
| image - words | -4.863889 | 0.1808043 | 72 | -26.901404 | 0.0e+00 |
| number - words | -1.000000 | 0.1808043 | 72 | -5.530843 | 1.4e-06 |
Our Fisher LSD test in Table 5 reveals statistically significant differences among all levels of our memory type factor. Specifically, in Table 6 the estimated contrast suggests that participants made significantly fewer errors in the image memory test compared to the number memory test (estimate = -3.863889, p < 0.001). Similarly, participants also made significantly fewer errors in the image memory test compared to the word memory test (estimate = -4.863889, p < 0.001). Lastly, participants made significantly fewer errors in the number memory test compared to the word memory test (estimate = -1.000000, p = 1.4e-06).
In conclusion, our analysis aimed to investigate the effect of fidget toys on memory performance. Contrary to claims made by fidget companies regarding the potential benefits of their toys in improving focus, our ANOVA results revealed no significant evidence to support such a relationship F(1, 0.368), p = 0.545. This suggests that the different fidget toys, namely poppit and fidget spinner, did not significantly impact individuals’ memory performance in our experimental study.
The interaction term between memory type and fidget was also not significant F(2, 0.150), p = 0.861. This indicates that the influence of fidget toy type on memory performance did not vary significantly across different memory tasks (image, numbers, and words). In other words, the effect of fidget toys on memory performance was consistent across all memory task types tested in our study.
We did, however, observe a significant main effect of memory type on the number of errors F(2, 402.115), p < 2e-16. This finding suggests that the type of memory test significantly influenced participants’ memory performance across the different memory tests. Post-hoc analyses revealed that participants tended to make more errors on certain types of memory tasks compared to others, with the biggest difference being between image and words memory test.
However, it’s crucial to interpret these findings with caution, as violations of additivity, standard deviations, and normality of residuals were observed. Additionally, attempts to rectify these violations through diagnostic transformation were unsuccessful. As such, future studies could use factorial design such that the interaction of fidget and memory type can be studied.
The limitations of our study include the relatively low sample size, which may have constrained the detection of significant correlations, despite being larger than the studies we referenced. Additionally, due to digital constraints, our investigation lacked the inclusion of a physical fidget toy, potentially impacting our findings on their effects. Furthermore, external factors such as environmental conditions or concurrent interventions from participants not under our control were not accounted for, possibly confounding the observed associations.
Future research could address these limitations by expanding the sample size, incorporating long-term physical interaction elements, and implementing more rigorous control measures for external factors.
Constant Effects
Constant effect is assumed to be reasonable, as all participants in each memory type level were exposed to the same stimuli images, words, numbers and GIF (either poppit or spinner).
Independence
It is reasonable to assume that the factors under consideration are independent, particularly as evidenced by correlation coefficients close to zero. A correlation coefficient near zero suggests a weak or negligible linear relationship between variables, supporting independence. Additionally, the Fisher LSD analysis further reinforces this independence, particularly regarding the factor of memory type. The observed grouping patterns and individual relationships with the number of errors indicate that memory type operates independently of other factors influencing error rates.
Additive Effects (Anscombe Plots)
a<- qplot(x = image, y = words, data = wider_df) +
geom_abline(intercept = 4, slope = 1, color = "blue", linetype = 2) +
geom_smooth(method = "lm", se = 0, color = "orange")
b<- qplot(x = image, y = number, data = wider_df) +
geom_abline(intercept = 2, slope = 1, color = "blue", linetype = 2) +
geom_smooth(method = "lm", se = 0, color = "orange")
c <- qplot(x = number, y = words, data = wider_df) +
geom_abline(intercept = 1, slope = 1, color = "blue", linetype = 2) +
geom_smooth(method = "lm", se = 0, color = "orange")
(a)/(b)/(c)
Based on our Anscombe Plots, it’s evident that the additive assumption is violated, as the points in the Anscombe block plots deviate from a slope of 1. This observation strongly indicates that an additive model is not suitable for our data.
Standard Deviations
# SD Check for both Memory and Fidget
s_condition_memory <- ds_memory |>
summarize(max(memory_sd)/min(memory_sd)) # S Assumption Violated!!!
kable(s_condition_memory)| max(memory_sd)/min(memory_sd) |
|---|
| 2.072533 |
s_condition_fidget <- ds_fidget |>
summarize(max(fidget_sd)/min(fidget_sd)) # S Assumption PASS!!
kable(s_condition_fidget)| max(fidget_sd)/min(fidget_sd) |
|---|
| 1.024882 |
For memory type there is a difference in standard deviations, with the maximum being more than twice the smallest (2.07), indicates inconsistency across groups.This suggests a need, for some kind of transformation. For fidget type, the standard deviation assumption is satisfied.
Normally Distributed Residuals:
## Normality is violated
plot(model1, which = 2)
In analyzing the data, it was observed that the QQ plot of residuals displayed violations of the Normality assumption. This suggests that the residuals are not normally distributed, which may impact the validity of the ANOVA results.
Zero Mean Residuals
qplot(model1$residuals, bins = 5)
## Distribution is rather normal, with residuals centered on zero, thus zero mean residual conditon is passed.The histogram of residuals suggests that the assumption of zero mean residuals is satisfied. In ANOVA, it is assumed that the residuals have a mean of zero.
Legacy Toys. (2023). Fidgets & Desk Toys. https://legacytoys.com/collections/fidget-toys↩︎
Rohrberger, A. (2011). The efficacy of fidget toys in a school setting for children with attention difficulties and hyperactivity (Master’s thesis). Ithaca College. https://core.ac.uk/download/pdf/217299391.pdf↩︎
Croley, K. E., Drevon, D. D., Decker, D. M., Hixson, M. D., & Radley, K. C. (2023). The Effect of the Fidget Cube on Classroom Behavior among Students with Perceived Attention Difficulties. Behavior Analysis in Practice, 16(2), 547–557. doi: 10.1007/s40617-022-00734-4↩︎
Kirby, T. (2020). The effect of fidget spinners on attention and anxiety in typically developing children (Master’s thesis). Washburn University. https://wuir.washburn.edu/server/api/core/bitstreams/58cd41fb-81eb-4754-a583-71f0d4ae549b/content↩︎