1. Introduction
It is not uncommon near the beginning of the winter season in parts of North America where snow is a regular occurrence to hear residents and the local news quip that people in their city, region, or state always “forget how to drive” when the first snow of the season falls (Canadian Broadcasting Corporation 2017; Hare 2015; Schultz 2014). Residents of Indiana, the so-called Hoosier state, are no exception (Mack 2019; Watson and Wilkins 2020). From a climatological perspective, different parts of Indiana can expect very different amounts of snowfall in a normal year: from as little as 27.4 cm (10.8 in.) in the Evansville area in the southwest to as much as 163.8 cm (64.5 in.) in the South Bend area in the northwest, with 85.3 (33.6 in.) and 64.8 cm (25.5 in.) being the norms in the Fort Wayne (northeast) and Indianapolis (central) areas, respectively (NOAA 2021). Not only do residents have to navigate the roadways when snow falls, but numerous travelers are often passing through the state on one of several interstate highways, supporting the state’s motto of “Crossroads of America,” which was adopted in 1937 (Indiana Historical Bureau 2021). This study addresses the question of whether there are more motor vehicle crashes in Indiana on the day of the first snowfall each winter season when compared with all other days with snowfall. In addition, differences in the number of injuries and fatalities between the first snow day and subsequent snow days was examined.
Weather, particularly snow, has been shown to have an effect on many aspects of driving, such as reductions in traffic volume (Call 2011; Burow and Atkinson 2019; Call and Flynt 2022). In two studies of traffic on the New York State Thruway, Call (2011) and Call and Flynt (2022) found that a negative correlation exists between traffic count and snowfall rates, with the largest impact associated with passenger cars. Similarly, Burow and Atkinson (2019) calculated a 10% reduction in traffic in northeast Ohio during select snow events. Snowfall not only affects traffic volume, but also plays a role in the rate of crashes during snow events (Burow and Cantrell 2021; Call and Flynt 2022). Call and Flynt (2022) determined that inclement weather, mainly snow, accounted for 42%–65% of crashes in January on the New York State Thruway. Further, snowfall rates were found to predict 30% of the variation in these crashes. Focusing on crashes in northern Indiana, Burow and Cantrell (2021) found that daily crash counts increased with increasing snowfall rates by about 3.1% cm−1 during light and moderate snow events; during heavy snow events (>17.3 cm in a day) crash counts were lower than the authors predicted by linear regression, likely because of fewer people attempting to drive.
Pisano et al. (2008) stated that weather affects driver capabilities, vehicle performance, and pavement friction, which can lead to motor vehicle crashes. They explain that at the beginning of winter drivers are presented with road conditions to which they are not accustomed, at least not recently, which results in higher crash rates earlier in the season. Since drivers perceive snow to be more dangerous than rain, they adjust their driving habits. Although crash rates may be higher, Tefft (2016) found that crashes during adverse weather or road conditions resulted in fewer fatalities relative to crashes during clear weather and dry roads. The author attributed this to people driving slower and more carefully when conditions are poorer. Although crashes, especially single-vehicle crashes, increase during snow events, there were 47% fewer fatalities per crash. In their study of vehicle-related fatalities during active precipitation events, Tobin et al. (2019) found that 14% of fatalities occurred during snowfall, with rain accounting for the majority of precipitation-related fatalities.
As part of their study on snowfall and motor vehicle collisions across the United States, Eisenberg and Warner (2005) identified the first snowfall of the season, defining it as the first day in each state with at least 0.5 cm of snow accumulation after 100 consecutive days without snow. They found that, relative to nonfirst snow days, there were more fatalities on first snow days, especially among drivers aged 65 and older. These authors also noted that serious outcomes from crashes were reduced on snow days relative to dry days, while less serious outcomes increased. Within their study on the effects of winter precipitation on automobile collisions, injuries, and fatalities, Black and Mote (2015) assessed the relative risk of crashes and injuries from the first snowfall of the season in 11 cities across the United States. Using the first three winter precipitation events in comparison with subsequent events, they found that the relative risk of crashes and injuries were higher during the first three events of the season, although these results were not statistically significant.
This study also investigated the relationship between crashes and the first snowfall of each winter season; instead of generalizing across entire states or a selection of cities across the country, the focus here was on the entirety of a single state. Because snowfall is distributed unevenly across a region, measurements of snow accumulations from weather stations across Indiana were used to identify snow days at multiple locations rather than for the entire state. Crashes, injuries, and fatalities in the proximity of these locations were identified, allowing for the construction of a more robust dataset. Analyses focused on the comparison of crashes from the first snowfall of each winter season with other days with snow, as well as injuries and fatalities.
2. Data and methods
Vehicular crash data in Indiana are available from 2007 to 2020 through reports submitted to the Automated Reporting Information Exchange Systems (ARIES) and disseminated on the Indiana Data Hub (https://hub.mph.in.gov/). Numerous variables are provided within this dataset, but pertinent to this study are the collision identification number, collision date, county where the collision occurred, the latitude and longitude of the collision, number of injuries, and number of fatalities. In most cases (more than 87% of the 2 839 869 crashes) there were latitude and longitude coordinates available. For those crashes without coordinates, the latitude and longitude of the center of the county in which the crash occurred was substituted, with only 50 crashes having neither coordinates nor a county identified (these crashes were omitted from this study).
Snowfall data were obtained from the National Centers for Environmental Information Global Historical Climate Network daily (GHCNd) database (Menne et al. 2012). For each of the 1366 weather stations in Indiana that were operational for any portion of the 2007–20 period of study, days with any measurable snow accumulation (≥1 mm) were recorded. All of the stations had latitude and longitude coordinates marking their geographic locations.
It would be easy to identify the first snowfall of each year for the entire state of Indiana, but only a portion of the state may be affected, whereas other parts might experience their first snowfall on a different day. Further, by considering the state as a whole, there would only be 14 “first snow” days available for analysis, which would not make for a robust dataset. Thus, it is beneficial to divide the state into regions, allowing for better representation of the crashes associated with snowfall in the area, and a larger dataset of “first snow” days. To this end, a grid of 46 points was constructed within and in the vicinity of the state of Indiana, with a grid spacing of 0.5° in both latitude and longitude (Fig. 1). Both stations and crashes were assigned to the point to which they were closest geographically, effectively forming a grid of 0.5° by 0.5° cells.
(left) Locations of the grid of points used in this study (red circles), with interstate highways and county outlines included for reference. (right) Locations of all GHCNd stations used (points), with stations of the same color denoting those associated with the same point (open black circles), and the associated grid cells marked with thin, black lines.
Citation: Weather, Climate, and Society 14, 4; 10.1175/WCAS-D-22-0055.1
For each day during the study period, any station with measurable snowfall was used to calculate the mean snow accumulation in each cell, and all crashes, injuries, and fatalities were summed within the cell. The question of driving performance on the day of the first snowfall does not inherently imply an amount of accumulated snow that causes problems for drivers, so a set of thresholds were chosen for examination: 1, 13 (∼0.5 in.), 25 (∼1.0 in.), and 51 (∼2.0 in.) mm. Beginning on 1 July each year, the first day on which each of the above thresholds was exceeded in each cell was the first snowfall of the winter season, and the crashes during that day were recorded for that cell. Note that in a cell there could be a different day that is the first snowfall for each threshold, but it could also be the same day for each threshold (i.e., if the first measurable snowfall of the season exceeded 51 mm).
Mean number of daily crashes in each cell.
Citation: Weather, Climate, and Society 14, 4; 10.1175/WCAS-D-22-0055.1
3. Results and discussion
The distribution of normalized daily crashes on days without any snow accumulation is nearly normal, with a mean of −0.06 and standard deviation of 0.90 (Fig. 3). On days with measurable snowfall, both the mean (0.80) and standard deviation (1.68) are higher. Using a two-sample t test, a significant difference (p < 0.001) was found between the normalized daily crashes on days with and without snow. Even though both distributions are positively skewed, there are higher frequencies of days with more crashes when snowfall occurs relative to days with no snowfall. This is not unexpected, as even small amounts of accumulated snowfall on roadways can present hazards while driving.
Frequency of normalized daily crashes on days with measurable (≥1 mm) snow accumulation (dashed line) and days without snow accumulation (solid line).
Citation: Weather, Climate, and Society 14, 4; 10.1175/WCAS-D-22-0055.1
On days with measurable snow accumulation, the distributions of crashes on the first day of the season for each of the four snowfall accumulation thresholds was skewed positively and had higher frequencies of higher normalized daily crashes relative to the distributions of crashes for other days with snow at each threshold (Fig. 4). The mean number of crashes on the first day with snow each season was significantly higher than those on the other days with snow at each of the four thresholds (Table 1), with p values from two-sample t tests each exceeding the 0.1% significance level. For the lower two thresholds there were about 2 times as many crashes on the first snow day than on other days with snow, while the difference fell to around 1.5 times as many for the higher two thresholds. These ratios were calculated by dividing the mean normalized crashes on the first snow day by the value from the other days with snow. The idea that drivers in Indiana crash more often on the first day of the winter season with snowfall regardless of the accumulation amount used to quantify the first snowfall is supported by these results.
Frequency of normalized daily crashes for the first snowfall of each season (dashed lines) and all other snowfall (solid lines) for accumulations of at least (a) 1, (b) 13, (c) 25, and (d) 51 mm.
Citation: Weather, Climate, and Society 14, 4; 10.1175/WCAS-D-22-0055.1
Number of first snow days and other days with snow (aggregated from the 46 cells across Indiana), mean normalized daily crashes for each of the four accumulation thresholds, and the ratio of mean normalized crashes on the first snow day to other snow days. The p values from two-sample t tests comparing crashes on the first snow day of each season and other days with snow are shown for each accumulation threshold.
Mean daily crash values increased on the first day with snowfall between the 1- and 13-mm thresholds, lowered slightly at the 25-mm threshold, and lowered even more at the 51-mm threshold (Table 1). On the other hand, on the remaining days with snowfall, mean daily crash values increased with increasing snowfall thresholds, except between 25 and 51 mm, which remained the same. One possible explanation for this slight decline in crashes on the day of the first snowfall (although still over 1.5 standard deviations above the mean) using higher thresholds is that drivers may have already reacclimated to driving in snow from previous snowfalls that season that fell below the threshold (i.e., 51 mm). In fact, the date of the first snowfall is strongly related to the threshold chosen (Fig. 5). Using the analysis of variance (ANOVA) test, significant differences were found to exist between the distributions of the day-of-year of first snowfalls at the four thresholds (p < 0.001), with a Tukey test at a 95% familywise confidence level indicating significant differences between each of the pairs of distributions, each at a significance level exceeding 0.1%.
Boxplots showing the distribution of the dates of the first snowfall according to accumulation threshold.
Citation: Weather, Climate, and Society 14, 4; 10.1175/WCAS-D-22-0055.1
Expanding the analysis of crashes to compare injuries and fatalities on days with measurable snow accumulation with those without snow, it was found that, similar to crashes, there were more normalized injuries on days with snow (0.17) than without snow (−0.01), whereas there were fewer normalized fatalities on days with snow (−0.02) than without (0.00). Both of these results were statistically significant using two-sample t tests, with the significance level for injuries (p < 0.001) greater than that for fatalities (p = 0.010). While drivers were more likely to be involved in a crash in which an injury occurred during snow events, the lower likelihood of fatalities in snow events supports the findings of Tefft (2016).
As with crashes in general, there were significantly more injuries during the first snow day of the season relative to all other days with snowfall, regardless of the snowfall threshold used (Table 2). At the lower two thresholds the difference was more pronounced, with about 3 times as many injuries during the first snowfall and 2 times as many at the higher two thresholds. In the same way as was done for crashes, ratios were calculated by dividing the mean normalized injuries on the first snow day by the value from the other days with snow. These results are similar to those of Black and Mote (2015), although, unlike in their study, the differences here are statistically significant. Also, as with crashes, there was a peak at the 13-mm threshold for the daily normalized injuries on first snow days, although in this case the decrease at the 25-mm threshold was larger.
As in Table 1 but for injuries.
The analysis of fatalities that resulted from crashes during the first snow of the season yielded results similar to those of Eisenberg and Warner (2005), with more fatalities occurring during the first snowfall at each of the four thresholds relative to subsequent snowfalls (Table 3). In this case, however, the only statistically significant result from a two-sample t test was at the 13-mm threshold (p = 0.068). An important consideration when interpreting the fatality results is the relatively low sample sizes across the four snowfall thresholds in comparison with crashes and injuries. This is to be expected, since only 5% of all crashes in this dataset resulted in injuries, and 0.3% resulted in fatalities; those numbers fall to 4% and 0.2%, respectively, when considering crashes only on days with snowfall.
4. Concluding remarks
The purpose of this study was to determine if there is any truth to the popular sentiment that drivers forget how to drive in snow each year. This was accomplished by identifying the first snowfall of each winter season at each of 46 grid cells across Indiana and summing the crashes, injuries, and fatalities in each cell on the same day; the same was done with all other snowfall days after the first. Results of statistical tests showed that there were significantly more crashes and injuries on the first day with snowfall each year relative to the number of crashes on subsequent days with snowfall, and while there were more fatalities on the first snow day, this was only significant at the 13-mm snowfall threshold. In addition, it was found that with increasing snow accumulation thresholds the first snow day occurred later in the winter season.
The finding that more crashes occur on the first snow day of the season supports the work of Eisenberg and Warner (2005). While their approach generalized snowfall as a binary event for each state and used a single threshold of 5 mm (although they stated they experimented with thresholds up to 20 mm with similar results), this study focused on a single state with crash numbers related more to locations that experienced snow and tested the relationship across a broad range of accumulation amounts. The latter difference yielded an interesting result: crashes on the first snow day decrease with increasing accumulation thresholds. This, too, supports findings from a prior study, where Burow and Cantrell (2021) noted that their linear regression model predicted crashes well with light and moderate snowfall, but crashes during heavy snow were less than predicted. While 51 mm is not suggested to be heavy, the decrease in crashes on first snow days with increasing accumulation thresholds may be attributed to similar reasons (i.e., more people staying home rather than driving).
Although the study by Black and Mote (2015) used the first three snow days of a season instead of just one and also focused on a selection of 11 cities across the United States, the results from this study were similar to theirs: more crashes and injuries occurred on the first day with snowfall in a winter season relative to subsequent snow days. Whereas these results from Black and Mote (2015) were not statistically significant, the differences in crashes and injuries in the present study were. It is difficult to speculate on why this is the case, but one possibility could be the gridded approach used in this study, which greatly increased the chances of detecting smaller-scale snow events, and the crashes associated with them.
Since Eisenberg and Warner (2005) found similar results using all 48 contiguous states, it is likely that the results from this study apply to most states with appreciable annual snowfall. Future work could take a similar approach as in this study but apply it to a state that experiences snow events more frequently, or one with a longer snow season. It is possible that such states may see a diminished effect, although it would be surprising for there still not to be a significant difference in crashes or injuries.
Acknowledgments.
This work utilized Ball State University’s beowulf cluster, which is supported by the National Science Foundation (MRI-1726017) and Ball State University, Muncie, Indiana. The constructive comments and suggestions made by three anonymous reviewers helped to improve the paper.
Data availability statement.
Data analyzed in this study were existing data that are openly available from the locations cited in the text and reference section.
REFERENCES
Black, A. W., and T. L. Mote, 2015: Effects of winter precipitation on automobile collisions, injuries, and fatalities in the United States. J. Transp. Geogr., 48, 165–175, https://doi.org/10.1016/j.jtrangeo.2015.09.007.
Burow, D., and C. Atkinson, 2019: An examination of traffic volume during snow events in northeast Ohio. Nat. Hazards, 99, 1179–1189, https://doi.org/10.1007/s11069-019-03786-y.
Burow, D., and H. M. Cantrell, 2021: Surface and radar observations of winter weather conditions during traffic crashes in northern Indiana. Prof. Geogr., 73, 392–403, https://doi.org/10.1080/00330124.2021.1895852.
Call, D. A., 2011: The effect of snow on traffic counts in western New York State. Wea. Climate Soc., 3, 71–75, https://doi.org/10.1175/WCAS-D-10-05008.1.
Call, D. A., and G. A. Flynt, 2022: The impact of snowfall on crashes, traffic volume, and revenue on the New York State Thruway. Wea. Climate Soc., 14, 131–141, https://doi.org/10.1175/WCAS-D-21-0074.1.
Canadian Broadcasting Corporation, 2017: “Did everyone forget how to drive in the winter?”: Every driver every year. CBC News, 3 November, https://www.cbc.ca/news/canada/edmonton/edmonton-first-winter-commute-1.4384656.
Eisenberg, D., and K. E. Warner, 2005: Effects of snowfalls on motor vehicle collisions, injuries, and fatalities. Amer. J. Public Health, 95, 120–124, https://doi.org/10.2105/AJPH.2004.048926.
Hare, M., 2015: Why do we forget how to drive in winter? Democrat and Chronicle, 28 December, https://www.democratandchronicle.com/story/news/2015/12/28/winter-driving-amnesia-rochester/77979298/.
Indiana Historical Bureau, 2021: Emblems & symbols. Accessed 25 November 2021, https://www.in.gov/history/about-indiana-history-and-trivia/emblems-and-symbols/.
Mack, J. L., 2019: For the Hoosiers who forgot, here’s how to drive in snow without being a dangerous jerk. IndyStar, 11 November, https://www.indystar.com/story/news/2019/11/11/tips-safely-driving-snow-hoosiers-who-forgot/2562835001/.
Menne, M. J., I. Durre, R. S. Vose, B. E. Gleason, and T. G. Houston, 2012: An overview of the Global Historical Climatology Network-Daily database. J. Atmos. Oceanic Technol., 29, 897–910, https://doi.org/10.1175/JTECH-D-11-00103.1.
NOAA, 2021: NOAA online weather data. Accessed 26 November 2021, https://www.weather.gov/media/climateservices/NOWData.pdf.
Pisano, P. A., L. C. Goodwin, and M. A. Rossetti, 2008: US highway crashes in adverse road weather conditions. 24th Conf. on Interactive Information Processing Systems (IIPS), New Orleans, LA, Amer. Meteor. Soc., 8.1, https://ams.confex.com/ams/pdfpapers/133554.pdf.
Schultz, F., 2014: Do people just forget how to drive on snow? Oshkosh Northwestern, 29 November, https://www.thenorthwestern.com/story/news/local/2014/11/29/people-just-forget-drive-snow/19666759/.
Tefft, B. C., 2016: Motor vehicle crashes, injuries, and deaths in relation to weather conditions, United States, 2010–2014. AAA Foundation for Traffic Safety Doc., 14 pp., https://aaafoundation.org/wp-content/uploads/2016/01/weather_Fact-Sheet_2016.pdf.
Tobin, D. M., M. R. Kumjian, and A. W. Black, 2019: Characteristics of recent vehicle-related fatalities during active precipitation in the United States. Wea. Climate Soc., 11, 935–952, https://doi.org/10.1175/WCAS-D-18-0110.1.
Watson, D., and R. Wilkins, 2020: Overnight snow produces slippery roads, multiple slide-off accidents. Journal & Courier, 16 December, https://www.jconline.com/story/news/2020/12/16/snow-produces-slippery-lafayette-area-roadways/3920734001/.
