It is estimated that 28,000 people were injured in crashes involving mobile phone use in 2019 in the United States (National Highway Traffic Safety Administration, 2021). Despite its obvious risk, mobile phone use while driving is still ubiquitous. An observational study revealed that 23% of drivers were using their mobile phones and 9% of them were texting (Kruger et al., 2018). Texting while driving (TWD) is a particularly dangerous form of distracted driving because it involves visual (e.g., looking away from the roadway), manual (e.g., taking a hand off the steering wheel), and cognitive (e.g., thinking about something other than driving) forms of distraction (Sherin et al., 2014). A study utilizing naturalistic observations with on-board cameras revealed that TWD led to 3.87 times more crashes or near-crashes among novice drivers (Klauer et al., 2014; see also Caird et al., 2008, 2014; Simmons et al., 2016; Stavrinos et al., 2018, for meta-analyses on effect of cell-phone use on driving outcomes).

Transportation researchers have identified various situational factors associated with TWD. These include: (a) when the distance to the destination is farther (e.g., Hayashi et al., 2016); (b) while stopped at a red light (e.g., Bernstein & Bernstein, 2015) or while driving at slower speed (e.g., Oviedo-Trespalacios et al., 2017); (c) while driving under intense road conditions (e.g., Atchley et al., 2011) or when the chance of motor-vehicle crashes is higher (e.g., Hayashi, Fessler, et al., 2018); (d) when the amount of fine for TWD is higher (e.g., Hayashi, Friedel, et al., 2019a); and (e) when the message received while driving is very important (e.g., Foreman et al., 2021) or the social relationship to the sender is closer (Foreman et al., 2019).

One noteworthy finding in the literature is that perceived risk of motor-vehicle crashes due to TWD is not a reliable predictor of TWD. Some studies have found that perceived risk is a very weak predictor of TWD (Atchley et al., 2011), a predictor of TWD in females but not in males (Struckman-Johnson et al., 2015), or a predictor of intentions to engage in TWD (Brown et al., 2019). A majority of studies, however, have found that higher perceived risk is not associated with lower engagement or intentions of TWD (e.g., Berenbaum et al., 2019; Oviedo-Trespalacios, King, et al., 2017; Rupp et al., 2016; Sullman et al., 2018), which is consistent with the finding that drivers engage in TWD despite being aware of its danger (e.g., Atchley et al., 2011; Harrison, 2011; Hayashi et al., 2015; Terry & Terry, 2016).

To better understand the discrepant findings in the literature, the present study further explored the relationship between perceived risk of TWD and engagement in TWD in college students (one of the main target populations for the problem of TWD; Feldman et al., 2011). If the discrepant findings on the relation between perceived risk and TWD stemmed from the heterogeneity of perceived risk among drivers who frequently engage in TWD, it should be that some drivers who frequently TWD perceive TWD as very risky, whereas others perceive TWD as not so risky. To test this possibility, this study proposes that cluster analysis is useful.

Cluster analysis refers to grouping entities based on their similarities in terms of selected variables, such that members of the resulting groups are homogeneous to other members of the same group yet heterogeneous to those of the different groups (Clatworthy et al., 2005). An important advantage of cluster analysis is that classification can be performed objectively by assigning the variables equal numeric weights to help minimize a priori bias (Haldar et al., 2008). In this study a hierarchical cluster analysis was conducted, which involves a series of sequential steps in which individual cases are merged together one at a time based on their similarities (Clatworthy et al., 2005; Yim & Ramdeen, 2015). The primary purpose of the present study was to provide a proof-of-concept of usefulness of hierarchical cluster analysis for identifying distinct and practically meaningful subgroups of drivers who differed in their perceived risk and frequency of TWD. It was hypothesized that the hierarchical cluster analysis would result in the identification of distinct subgroups of drivers who would differ significantly in their perceived risk and frequency of TWD.

If such subgroups of drivers are identified, a next logical step is to evaluate the meaningfulness of the identified subgroups by investigating what factor(s) would differentiate them. To this end, it is important to recall the impulsive nature of TWD that drivers engage in TWD despite being aware of its danger (e.g., Atchley et al., 2011). Consistent with this, various studies (e.g., Hayashi et al., 2015; Pearson et al., 2013; Struckman-Johnson et al., 2015; but see Hayashi et al., 2016) have found that higher frequencies of TWD are significantly associated with higher levels of trait impulsivity (i.e., a predisposition toward rapid, unplanned reactions despite their negative consequences; Moeller et al., 2001) as well as impulsive decision making (i.e., a propensity to forego a future but larger reward to obtain an immediate but smaller reward; Green & Myerson, 2004). Based on these studies, the present study further explored whether subgroups of drivers who differ in their perceived risk and frequency of TWD would differ in terms of trait impulsivity and impulsive decision making. In addition, given the gender difference observed in the relationship between trait impulsivity and TWD (Struckman-Johnson et al., 2015), as well as the relationship between impulsive decision making and other problematic mobile phone use (e.g., Blessington & Hayashi, 2020), the comparisons among clusters were conducted separately for each gender. Because these were exploratory investigations based on the results of the cluster analysis, the study had no hypotheses to test for these analyses.

Fig. 1 shows the coefficient values of the cluster analysis as a function of the number of clusters. There was a robust increase in coefficient values (i.e., “jump”) as the analysis shifted between the model with three clusters and the one with two clusters, suggesting that the model with three clusters best fits the present data. Fig. 2 shows perceived risk and frequency of TWD of each participant as a function of the clusters. Overall, the three clusters show distinct patterns in terms of the scores of perceived risk and frequency of TWD. Cluster 1 was characterized by high risk and relatively high frequency (hereafter, High-Risk-High-Frequency subgroup). Cluster 2 was characterized by high risk and low frequency (High-Risk-Low-Frequency subgroup). Cluster 3 was characterized by relatively low risk and relatively high frequency (Low-Risk-High-Frequency subgroup).

Table 1 shows the demographic characteristics for the three clusters. There was a statistically significant difference across the clusters for gender, χ²(2) = 6.14, p =.046, but no statistically significant differences were found for age, F(2, 133) = 0.62, p =.542; years of higher education, F(2, 133) = 1.26, p =.287; and years driving, F(2, 133) = 0.16, p =.856.

Table 2 shows the results of the ANOVA on the BIS and MCQ scores conducted separately for each gender. For females, no significant differences among the clusters were observed on all measures, whereas for males, significant differences were observed on the BIS motor impulsivity and total scores. The results of the post-hoc comparisons (Table 3) revealed that for both measures, the High-Risk-High-Frequency subgroup showed significantly greater impulsivity than the High-Risk-Low-Frequency and Low-Risk-High-Frequency subgroups (p’s < 0.05).

Lastly, to confirm that splitting each cluster based on gender did not invalidate the clustering, gender differences on perceived risk and frequency of TWD were analyzed for each cluster. The results revealed that there were no gender differences on both measures for all clusters (p’s > 0.05).

Using a hierarchical cluster analysis, the present study explored whether college students could be categorized into distinct subgroups based on their perceived risk and frequency of TWD. The results revealed the following distinct subgroups: (a) drivers who perceive TWD as risky but frequently engage in TWD; (b) drivers who perceive TWD as risky and infrequently engage in TWD; and (c) drivers who perceive TWD as not so risky and frequently engage in TWD. To the best of our knowledge, this is the first demonstration that drivers who frequently engage in TWD can be categorized into two distinct subgroups that differ in terms of the perceived risk of TWD, suggesting that hierarchical cluster analysis is a useful research tool for identifying and profiling distinct, practically meaningful subgroups of drivers.

We believe this finding could potentially reconcile the discrepant findings on the relationship between perceived risk and frequency of TWD (e.g., Atchley et al., 2011 vs. Oviedo-Trespalacios, King, et al., 2017). Simply put, if a sample has a higher proportion of drivers who perceive TWD as risky but frequently engage in TWD, then it becomes more difficult to detect the significant negative relationship between perceived risk and frequency of TWD. Given the hypothetical nature of this argument, however, it is advisable for future research to further examine this hypothesis.

Another important finding of the present study is that male drivers, but not female drivers, who perceive TWD as risky but frequently engage in TWD showed significantly higher levels of trait impulsivity—but not impulsive decision making—than drivers who perceive TWD as risky and infrequently engage in TWD or those who perceive TWD as not so risky and frequently engage in TWD. This finding has three important implications. First, the differences between genders, as well as between trait impulsivity and impulsive decision making, may be relevant to the results of a meta-analysis that significant sex differences were observed with trait impulsivity (i.e., males being more impulsive than females) but not with impulsive decision making (Cross et al., 2011). Second, the gender moderation in the relationship between trait impulsivity and TWD in the present study would suggest that the significant positive relationship between trait impulsivity and TWD frequency reported in the literature (e.g., Pearson et al., 2013) is likely to be driven by male impulsive drivers who frequently engage in TWD. Third, the gender moderation would also suggest that intervention strategies targeting the impulsive nature of TWD may need to be trailed for each gender (details discussed follow).