If you begin sorting through state and national accident data, you'll quickly realize that it is nearly impossible to compile a 100% accurate assessment of the impact of icy roads in terms of injuries, fatalities and total accidents. The issue is a complex one, stemming mostly from the fact that our official systems of recording crashes are imperfect. State and national vehicle crash data are sourced from the reports filled out by officials on the scene of accidents. While the officials on the scene fill out the reports as best they can, incomplete or missing data is common. Reports may be corrected or added to in follow-up investigations, but many missing fields often remain in the final data.
This means it is challenging to assess the impact of icy roads on vehicle crashes. These are the three primary factors:
Many reports have missing/blank fields or data recorded as "unknown". One way that data can be incomplete is if the official filling out an accident report doesn't record ice, snow or slush in the road surface condition field(s). If this occurs, the incident will go into the state and national databases without any indication that icy road conditions were present at the time of the crash.
This chart of daily accidents in 2013 in Missouri shows this issue prominently. That year, a rare May snowstorm coated roads in the state, resulting in a large number of accidents on the 2nd and 3rd. However, on May 3, only 148 of the nearly 800 accidents that day were officially recorded with the road surface condition marked as "snow" or "slush". The large spike in accident numbers for that day clearly marks the snow event, but the majority of the crashes were not attributed to the snow in final accident reports:
There are also examples of high-profile accidents reported in news media as ice-related in which the final official report did not record an ice, snow or slush road surface condition . In such instances, there is no way to assess the true role of the road conditions without doing further detailed investigation of each individual accident. This could include researching insurance claims and interviewing accident victims, witnesses and first responders. Such an effort would be expensive, time consuming and highly impractical.
Road surface condition is not the only data that sometimes is not present in accident reports. Examples of other fields missing or marked as "other/unknown" in some crash reports include seat belt use, occupant age/gender, vehicle type and injury severity. This missing data means that some of the metrics in the icyroadsafety.com reports may not add up to equal the total number of crashes in the database. The reports on this site simply display the information that is available in the crash databases "as-is".
During high-volume events, many accidents are not recorded at all. The second factor that results in incomplete data is that during winter weather events with high accident rates, first responders and law enforcement can become overwhelmed. In such cases, many accidents will simply go unrecorded in the official data. I have examples of significant accidents that I witnessed and captured on camera [2, 3] that are not in any of the official databases. In those cases, the events were characterized by a very large number of crashes in a short period of time. There was simply not enough available personnel to properly fill out reports on all of the crashes. This results in the actual number of accidents being under-reported in official data for those events.
Some accidents on ice/snow might not have been caused by the road condition. The "road surface condition" field in crash reports is describing the condition present at the time and place of each accident. It does not state whether or not the conditions were either the cause of or a significant contributing factor to the crash. In northern states where snow and ice is present on many streets and roads for long periods of time, "everyday" accidents unrelated to the road conditions will nonetheless be included in these totals. There is no way to filter these out unless the data in the report explicitly notes a cause other than the road condition.
This issue is complex and unlikely to be resolved without detailed investigations into each individual crash to determine whether the road conditions were the cause or at the very least a conributing factor in the event. Absent this information, we simply don't have a practical way to determine which specific accidents to count and which ones to toss out.
All this being said, we can look at charts that show daily crash totals and plainly view the increases in the accidents when ice and snow are present. This example from Missouri in 2016 is illustrative:
The data shows that Missouri's daily crash totals average around 400 per day. The spike in raw totals during winter weather events is very clear. While we know that some of the accidents on ice/snow/slush covered roads (blue bars in the chart) may not have necessarily been caused by the road condition, it is clear by the totals alone that the mere presence of these conditions correlate with significantly higher accident rates.
Accidents involving deaths and serious injuries are typically more thoroughly investigated and recorded, so it appears discrepancies in death and injury counts are much less prominent. They still exist, however, as I have found several examples of such in the course of this research.
The takeaway is that the figures derived from these reports are not exact. There are factors present that both increase and decrease the number of accidents counted as snow and ice related. It is unclear exactly how large any discrepancy is - it can never truly be known without a further detailed - and very expensive - investigation of every single crash.