By law, all police agencies in the state are required to complete and submit to MSP a standard traffic crash report form. Nearly all agencies in Michigan electronically submit their UD reports to MSP, but a small number still submit hard copies. MSP provided traffic crash data for the first six months of However, we restricted the traffic stop data to the same time of year that the traffic crash represents, which helps account for seasonal fluctuations in the data. Moving forward, MSP will have the ability to compare traffic stop and crash data contemporaneously.
We restricted our analysis to the , motor vehicle drivers involved in traffic crashes in Michigan during the observation period. Accordingly, our analysis did not consider data from bicyclists, pedestrians, passengers, or train engineers involved in crashes.
As such, we are missing traffic crash data from a non-trivial number of Michigan police agencies. This is a significant limitation to the traffic crash benchmark that we will discuss more later. The data represent traffic crashes reported by law enforcement agencies in Michigan. We coded. This removed 33, non-drivers from the analysis. Table 4. State of , , 27, 3, 1, 75, 56, 59, 73, Michigan District 1 16, 13, 2, 29 9, 6, 24 7, 9, Accordingly, a larger percentage of drivers involved in crashes are African American than would be expected based on their composition in the population e.
There could be many reasons why this difference exists that the data cannot address. For example, African Americans may be more likely than their counterparts to drive in heavily populated and congested areas compared to White drivers across Michigan.
This could increase their exposure to traffic crash risk and disparate outcomes related to traffic stops. It is important to emphasize that these traffic crash reports were completed by all police agencies in Michigan and, thus, do not only represent those investigated by MSP. In most cases apart from Districts 7 and 8 , these percentages are about 1. About However, about At the district level, For example, Of the crashes in Allegan, Kent, Ottawa,.
Restricting our focus to only crashes investigated by MSP would produce an arbitrary and unrepresentative sample of the driving population. Table 5. State of Washtenaw County had the highest percentage of crashes involving Asian drivers 2. To do so, we restricted our attention to only those traffic stops that occurred during the intertwilight period. This is the period between the earliest end of civil twilight and the latest end of civil twilight.
We also omitted stops that occurred during the roughly minute period between sunset and the end of civil twilight i. Daylight is classified as clock times that fall before sunset. Darkness is defined as clock times falling after the end of civil twilight. This method provides a natural experiment of sorts because some clock times during the intertwilight period are daylight during one time of year but dark during another.
The method required us to calculate sunset and civil twilight times within the intertwilight period for each county in Michigan.
Sun times were calculated utilizing the suncalc package for RStudio github. The statistical package derives times based on the position of the sun and Earth www. We used this information along with the latitude and longitude of the center of each Michigan county to derive county-specific sunset and civil twilight times.
Accordingly, the earliest and latest end of civil twilight times varied by county but ranged from pm to pm. Restricting our analysis to this intertwilight period within each county resulted in 68, traffic stops included in the VOD analysis. Second, we benchmarked the traffic stop data with the traffic crash data. The results for the Census and traffic-crash benchmark analyses are presented in tables that feature several pieces of information. Next, the tables report z-statistics, which are tests of whether two population means are different from one another in a statistically meaningful way.
For example, a statistically significant z-statistic would tell us that the percentage of African-American drivers stopped by MSP is different either. The tables also provide an odds ratio for each comparison which is useful for interpretation purposes. The third section of the results focused on the VOD hypothesis.
We provide more information about this analysis below. The final section of the results centered attention on the stop outcome analysis. Here, again, we used logistic regression to predict the odds of a driver receiving particular outcomes after the stop is initiated e. More details on the analytic strategy are provided below. Census Benchmark Table 6 compares the distribution of traffic stops conducted by MSP troopers in involving an African-American driver to the distribution of African Americans in the population according to Census estimates.
Before presenting these results, it is important to underscore that relying solely on Census data benchmark results to assess the extent of racial disparity in traffic stops is inappropriate, potentially misleading, and entirely insufficient to address whether discrimination or bias exists in MSP trooper traffic stop behavior. An odds ratio of 1. Odds ratios can also be interpreted in a different way. Take, for example, an odds ratio of 3. Table 6.
Comparison of African-American traffic stops to African-American representation in population. Statewide Red highlighting indicates that the percentage of stops involving African-American drivers is higher than would be expected based on their representation in the population. Green highlighting indicates that the percentage of stops involving African-American drivers is lower than would be expected based on their representation in the population. Gray highlighting indicates that the percentage of stops involving African-American drivers is consistent with what would be expected based on their representation in the population.
Across the entire state of Michigan, The difference between these percentages is statistically significant at the 0. The largest difference was observed in District 3. Within this district, African Americans were about three times more likely to be stopped by District 1, 2, and 5 troopers and about two times more likely to be stopped by District 6 and 7 troopers, respectively, based on their representation in the population.
Appendix D provides the same comparison at the county level. As the table reveals, most Michigan counties experienced the same pattern of disparity as discussed above. For example,. A z-statistic that is at the 0. Wayne County has the most diverse population in the state.
It is too tedious to go through every county presented in this table, but a few are worth pointing out. Livingston County, for example, has a very small African- American population 0. In Genesee County, It is important to note that an opposite pattern of results was observed in several counties. One of the big take-away points from Appendix D is that there is a lot of variation in traffic stop disparities across Michigan counties for African Americans when using Census data as a benchmark.
Here again, we see disparity in traffic stops but in the opposite direction and magnitude of the African-American disparity observed above. About 5. The same pattern of results remained across all MSP districts.
The percentage of stops involving Hispanic drivers was about what we would have expected in District 5 based on the ethnic composition of the population based on the non-statistically significant z-statistic. Appendix E presents these comparisons for Hispanic drivers across Michigan counties. Throughout most counties, Hispanic drivers were less likely to be pulled over by MSP troopers than expected based on the composition of the population.
In only one county did Hispanics face disparity in traffic stops. In Berrien County, 7. About 3. Table 7. Comparison of Hispanic traffic stops to Hispanic representation in population.
Statewide 2. District 1 2. Red highlighting indicates that the percentage of stops involving Hispanic drivers is higher than would be expected based on their representation in the population. Green highlighting indicates that the percentage of stops involving Hispanic drivers is lower than would be expected based on their representation in the population. Gray highlighting indicates that the percentage of stops involving Hispanic drivers is consistent with what would be expected based on their representation in the population.
More variation exists when we examined the Asian benchmark across counties. Appendix F presents these results, and it is clear from the non-statistically significant z-statistics for many counties, Asians were represented in traffic stops in a manner consistent with what we would have expected based on their population representation. However, there are also numerous counties where Asians were less likely to be stopped based on their population composition, and some counties where Asians were stopped at a rate greater than expected based on their population composition.
Census benchmark limitations Census data benchmarking is a good starting point, but it is problematic because it does not represent the driving population. It is possible and highly likely that the driving population is not necessarily the same as the residential population, especially in areas with interstate highways.
This limitation is particularly salient when examining traffic stops conducted by a state police agency with a great deal of presence on interstates.
Statewide 0. District 1 1. Red highlighting indicates that the percentage of stops involving Asian drivers is higher than would be expected based on their representation in the population. Green highlighting indicates that the percentage of stops involving Asian drivers is lower than would be expected based on their representation in the population. Gray highlighting indicates that the percentage of stops involving Asian drivers is consistent with what would be expected based on their representation in the population.
Traffic-crash benchmark Traffic crash data is a commonly used benchmark when examining traffic stop racial disparities because it does a better job estimating the driving population than does Census data.
This allowed us to compare the traffic stops to crashes during the same time of year. While we do not have crash data for , this strategy allows us to account for potential seasonal differences in stops and crashes throughout the year. For example, if we compared traffic stops for all of to the crash data for which we only have the first six months of , differences in stop activity from July through December could confound the benchmark analyses.
During this period, This is a meaningful, but moderate, level of racial disparity. For example, in District 6 African. This was the least amount of racial disparity out of the seven districts. District 3 had the largest gap where African-American drivers represented Table 9. Red highlighting indicates that the percentage of stops involving African-American drivers is higher than would be expected based on their representation in not-at-fault crashes.
Green highlighting indicates that the percentage of stops involving African-American drivers is lower than would be expected based on their representation in not-at- fault crashes. Gray highlighting indicates that the percentage of stops involving African-American drivers is consistent with what would be expected based on their representation in in not-at-fault crashes. Appendix G presents a county-level comparison of African-American driver representation in traffic stops and crashes.
Joseph, Tuscola, and Van Buren. Although Muskegon County has fewer traffic stops and crashes than more populated counties, it had the largest gap. As we mentioned earlier, however,. It is possible if not likely that large differences observed in Appendix G are partially caused by this lack of vital traffic crash data.
Red highlighting indicates that the percentage of stops involving Hispanic drivers is higher than would be expected based on their representation in not-at-fault crashes. Green highlighting indicates that the percentage of stops involving Hispanic drivers is lower than would be expected based on their representation in not-at-fault crashes. Gray highlighting indicates that the percentage of stops involving Hispanic drivers is consistent with what would be expected based on their representation in not-at-fault crashes.
The same results emerged in the county-level comparison presented in Appendix H. Appendix I presents the county-level comparison and the same results emerged. In only two counties were Asian drivers more likely to be stopped than we would have expected. Red highlighting indicates that the percentage of stops involving Asian drivers is higher than would be expected based on their representation in not-at-fault crashes.
Green highlighting indicates that the percentage of stops involving Asian drivers is lower than would be expected based on their representation in not-at-fault crashes. Gray highlighting indicates that the percentage of stops involving Asian drivers is consistent with what would be expected based on their representation in not-at-fault crashes.
The analyses discussed above focused on all traffic crashes regardless of the number of vehicles involved. The problem, however, is that the racial disparity research literature is unclear on whether this is the appropriate methodology.
The argument is that a driver being involved in a two-vehicle crash for which they are not responsible. Deficiencies in driver ability regardless of whether the driver is deemed at-fault or not are more likely to be a factor in single- vehicle crashes than two-vehicle crashes according to this view.
Table 9-Supplemental presents the results of this supplemental analysis. This caused the disparity that was originally observed statewide in Table 9 above to disappear. Table 9-Supplemental. Note: These comparisons are restricted to only two-vehicle crashes. Percentages may not sum to due to rounding. Red highlighting indicates that the percentage of stops involving African-American drivers is higher than would be expected based on their representation in not-at- fault crashes.
Green highlighting indicates that the percentage of stops involving African-American drivers is lower than would be expected based on their representation in not-at-fault crashes. The racial disparity patterns remained when we examined the comparisons across each of the MSP districts. We also re-ran the analyses across each Michigan county and those results are presented in Appendix G-Supplemental. The substantive results remained unchanged. Table 12 presents this comparison for African Americans.
However, in Districts 1, 2, 3, and 5 African- American Table Red highlighting indicates that the percentage of stops involving African-American drivers is higher than would be expected based on their representation in at-fault crashes. Green highlighting indicates that the percentage of stops involving African-American drivers is lower than would be expected based on their representation in at-fault crashes.
Gray highlighting indicates that the percentage of stops involving African-American drivers is consistent with what would be expected based on their representation in at-fault crashes. The county-level comparison results are presented in Appendix J. This suggests that part of the disparity in traffic stops could be attributed to different driving behavior among African Americans. These data cannot speak to which explanation is correct.
Red highlighting indicates that the percentage of stops involving Hispanic drivers is higher than would be expected based on their representation in at-fault crashes.
Green highlighting indicates that the percentage of stops involving Hispanic drivers is lower than would be expected based on their representation in at-fault crashes.
Gray highlighting indicates that the percentage of stops involving Hispanic drivers is consistent with what would be expected based on their representation in at-fault crashes. In Districts 2, 3, 5, 7, and 8, Hispanic drivers were stopped at a rate that we would have expected based on the benchmark. The same pattern of results emerged in the county-level comparison that is presented in Appendix K.
It is important to point out, however, that this is based on a relatively low base rate of Asian driver involvement in stops and crashes. When we do this comparison at the county level see Appendix L , there were zero counties in Michigan for which comparisons could be calculated where Asian drivers were more likely to be stopped than expected.
Red highlighting indicates that the percentage of stops involving Asian drivers is higher than would be expected based on their representation in at-fault crashes. Green highlighting indicates that the percentage of stops involving Asian drivers is lower than would be expected based on their representation in at-fault crashes.
Gray highlighting indicates that the percentage of stops involving Asian drivers is consistent with what would be expected based on their representation in at-fault crashes. There are disparities with respect to African-American drivers that warrant closer examination. At the same time, however, there are several limitations with these benchmarks that are worth noting. First, not all Michigan crashes were represented in the crash database. Accordingly, a portion of the racial disparities observed in these benchmark analyses may be attributable to missing data rather than problematic behavior on the part of MSP troopers.
If this were to occur, the amount of disparity currently attached to MSP traffic stop behavior would be reduced or eliminated. Of course, it is also possible that the opposite could occur. Unfortunately, we are unclear about the true racial makeup of Michigan traffic crashes due to this missing data.
Second, the traffic stop and crash data did not cover an entire year i. This introduces uncertainty about whether the same conclusions would be reached if we had complete data to work with. The good news is that MSP will be able to conduct such analyses in the future as data becomes available. Third, and relatedly, the stop and crash data came from different years i. While people worked from home more often during , this trend may not have applied equally across racial groups.
In particular, African Americans may have been on the road traveling to and from work more often than their White counterparts because, as a group, they may have been more likely to work in service-oriented jobs that could not operate remotely. If true, it is possible that a larger proportion of drivers on the road were African American in compared to normal years.
When driving patterns return to normal, it is possible the observed disparities will become weaker. Answers to such questions await future inquiry. The final limitation with the traffic crash benchmarks is that they are imprecise when it comes to matching the locations that troopers patrol. Police officers often are deployed in strategic manners.
If African Americans are more likely to live or work in areas with such concerns, we would have expected traffic stop disparities to a certain degree. Unfortunately, the traffic crash data benchmarks do not resolve this problem because they are aggregated to a level that is larger than assigned patrol locations in many situations.
Ideally, we would be able to geographically locate both traffic stop and crash data to create more precise comparisons. However, such data and analytic capabilities are often unavailable.
We also controlled for the day of the week Sunday was used as the reference category and the time the traffic stop took place. We created a time bin to represent the clock time of the traffic stop by dividing the intertwilight period into minute intervals. Following the approach used by Taniguchi et al.
The models also controlled for trooper assignment type general patrol assignment was used as the reference category and the county-level violent crime rate where the stop occurred. A majority of traffic stops during the intertwilight period involved a White driver The racial composition of drivers during daylight and darkness varied.
Among stops that occurred during daylight, However, The percentage of stops involving African-American drivers was slightly smaller during darkness. The first model examined whether daylight predicted whether a driver was African American. The results reveal a statistically significant relationship between driver race and daylight. Daylight traffic stop. Day of the week b Monday. Time bin c. County-level violent crime rate. Intercept Note: Entries are unstandardized partial regression coefficients b , robust standard errors that adjust for clustering at the county level SE , and odds ratios OR.
Several other variables were significantly associated with the odds of a driver being African American. Specifically, stops conducted later in the evening were more likely to involve African- American drivers, net of other factors accounted for in the model.
Trooper assignment emerged as a significant predictor of driver race. According to Taniguchi et al. DST started on March 8th in Therefore, this analysis considers MSP traffic stops that occurred from February 7, through April 6, This resulted in 9, stops in the restricted VOD analysis. The results from these analyses are presented in Appendix M. Two important results emerged that conflict with the main VOD analyses discussed earlier. First, with the DST restriction, daylight no longer predicted the odds of a driver being African American.
Second, the final model in Appendix M demonstrates that traffic stops conducted during daylight during the days before and after the change to DST were about 3.
These findings provide some caution when interpreting the VOD results. As Taniguchi et al. While it is a valuable technique, the inconsistent findings underscore the importance of not relying solely on the VOD methodology when assessing traffic stop disproportionality. Veil of darkness benchmark limitations In addition to the mixed VOD results, there was an important limitation related to the DST- restricted analyses.
Specifically, the observation period for these analyses February 7 through. It is possible that this caused the mixed results to emerge. However, it is also possible that the DST-restricted analyses produced valid results.
Unfortunately, our data cannot speak to which possibility is correct. Stop Outcome Analysis We now turn our attention to the post-stop portion of the analysis, where we explored the different outcomes that stemmed from MSP traffic stops. Our goal here was to examine whether African-American, Hispanic, or Asian drivers disproportionately received sanctions during traffic stops, net of other factors that may influence trooper decision making during post-stop activities.
For the stop outcome analysis, we used the same traffic stop, Census, and crime data as described above. However, we used several new variables for this portion of the analysis, each of which is described next. It is important to note that these categories were not mutually exclusive because multiple outcomes could have occurred during the same traffic stop. For example, during a single traffic stop, a trooper could issue a warning for a broken taillight and citation for speed.
During the same encounter, the trooper could have conducted a consent search and arrested the driver based on contraband found during that search. Searches were grouped into three categories based on the level of discretion available to the trooper—consent, high-discretion, and low-discretion searches.
Low-discretion searches included searches incident to a lawful arrest, vehicle inventories, or warrants. We only considered consent searches and high- discretion searches in the outcome analyses because trooper discretion is limited in the low- discretion searches.
MSP troopers list the reason for the traffic stop when completing their reports. There are hundreds of options that troopers can pick from when listing the reason for the stop.
Accordingly, we controlled for whether the. As discussed earlier, the structural features of the communities that troopers patrol may impact their post-stop decision-making.
Accordingly, we accounted for district- and county-level structural characteristics in their respective analyses. Concentrated disadvantage also has been shown to predict police officer post- stop behavior COPS, Accordingly, we accounted for the level of concentrated disadvantage in the district or county where the traffic stop took place. The variable is a mean scale constructed by summing and averaging values from three structural indicators: the percentage of the population that is unemployed, the percentage that lives below the poverty line, and the percentage of households that are female-headed.
Recall from our earlier discussion that we calculated the violent crime rate which represents the number of violent crimes in per , residents at the district- and county-level of analysis, respectively. Prior research has shown that the violent crime rate influences officer post-stop behavior COPS, Accordingly, the post-stop outcome analyses account for the violent crime rate of the location the stop took place.
Stop outcome analytic strategy The stop outcome analyses consisted of two steps. Specifically, a separate logistic model was estimated for each of the four stop outcomes. Logistic regression was used in this stage of the analysis because the dependent variables i. Each model accounted for the variables described above. However, it did not load with the other structural indicators i. We conducted two sets of analyses—one that controlled for district characteristics and another that controlled for county characteristics where the traffic stop took place.
We also used robust standard errors that adjusted for clustering at the district- of county-level in each set of analyses, respectively. About the same percentage of White and African-American drivers received a warning Likewise, Warning Citation Search Arrest. Total , 90, 16, 20, African American 48, 20, 7, 8, Hispanic 5, 2, Asian 1, 1, 36 43 Note: Entries represent the frequency and percentage of traffic stops that involved the respective outcome.
While instructive, these comparisons do not account for the reason for the stop or community characteristics where the stop took place. The logistic regression models presented in Table 18 do so. This table provides the results from logistic regression equations for each of the stop outcomes and that controlled for the district characteristics in which the stop took place. With respect to district characteristics, several findings emerged from the analysis worth mentioning.
We now turn our attention to Model 2 in Table 18 that presents the results of the equation that predicted the odds of receiving a citation. Here again, African-American and Hispanic drivers were no more likely to receive a citation than White drivers.
As expected, drivers stopped for a hazardous violation were about 3. The district-level characteristics were also associated with the odds of receiving a citation in a manner consistent with what we would have expected based on Model 1.
Asian -. Reason for stop Hazardous -. District characteristics Concentrated disadvantage. Violent crime rate -. Note: Entries are unstandardized regression coefficients b , robust standard errors adjusted for clustering in the seven MSP districts SE , and odds ratios OR.
African American represents non-Hispanic African Americans. Model 3 provides the results of the logistic model that predicted whether a driver was searched during the traffic stop. This may. After all, the purpose of many pretextual stops is to investigate other potentially dangerous criminal activity e.
It is important to underscore that the arrest may or may not have been related to the traffic stop itself e. Again, this seems to support the idea that non-hazardous violations are typically used as pretextual stops because they are more likely to lead to an arrest. This also suggests that a non- trivial portion of pretextual stops led to found contraband or people with warrants as evidenced by the increased odds of an arrest occurring. To summarize some of the key results from Table 18, African-American and Hispanic drivers were more likely to be searched and arrested after a traffic stop compared to White drivers after controlling for the reason for the stop and the district-level characteristics.
However, they were no more or less likely to be issued a warning or citation compared to White drivers. Asian drivers were more likely to be issued a citation, but less likely to be searched or arrested than White drivers.
The reason for the stop was important across each model. Drivers stopped for hazardous violations were more likely to be issued a citation but less likely to be given a warning, searched, or arrested. Concentrated disadvantage was also associated with each of the post-stop outcomes.
Drivers stopped in districts with greater disadvantage were less likely to be given a citation. This could have occurred for a number of reasons. Perhaps troopers have more problems to deal with in disadvantaged districts and are more likely to be lenient on drivers. Or, it is possible that more pretextual stops occur in disadvantaged districts which decreases the likelihood of issuing a citation.
Given that concentrated disadvantage was positively associated with the odds of a search and negatively associated with the odds of an arrest, this possibility is worth pursuing in greater detail with an analysis capable of considering a more fine-grained measurement of the reason for.
Drivers stopped in districts with larger African-American populations were more likely to receive a citation and be arrested, but less likely to be searched or given a warning. Drivers stopped in districts with larger Hispanic populations were less likely to receive a citation but more likely to be searched and given a warning, all else equal. County characteristics Concentrated disadvantage.
Violent crime rate. Note: Entries are unstandardized regression coefficients b , robust standard errors adjusted for clustering in the 83 Michigan counties SE , and odds ratios OR. Table 19 presents results from the same post-stop outcome logistic models discussed above, but this time we controlled for county-level characteristics.
While there were slight differences in the size of some of the effects, the results are virtually identical to those presented in Table 18 that controlled for district-level characteristics. Post-stop outcome analyses limitations We were able to account for the reason for the search and aggregate-level violent crime rates along with other important variables in the search and arrest post-stop outcome analyses.
However, it would be helpful to consider the prior criminal history of a driver when conducted these analyses. The SCP involves cooperation with several local police departments around Michigan where MSP provides patrol support to assist with violent crime problems.
This allowed us to focus on only those traffic stops that occurred in an SCP location and were related to SCP activities rather than including stops that happened to occur in one of the cities but were not part of SCP activities.
Table 20 provides the descriptive statistics for the SCP traffic stops that occurred in Only This is at least partially expected given that many of the SCP locations have higher percentages of African-American residents than the larger counties they are part of.
A majority of SCP stops occurred in two cities— Flint Table 21 provides the results from a benchmark comparison of the. Driver Gender Male 13, The table reveals that, indeed, a larger percentage of the population in the SCP locations is African American compared to the counties, districts, and statewide results described earlier.
However, a significant amount of disparity existed in the SCP locations. This finding holds true in eight of the SCP cities. The percentage of African-American drivers stopped by MSP troopers in Harper Woods, Highland Park, and Pontiac was what we would have expected based on their representation in the respective city populations.
It is important to note, however, that there were relatively few SCP-related stops in these locations. All SCP Locations SCP traffic-crash benchmark Next, we benchmarked the percentage of African-American drivers involved in traffic stops to the racial composition of not-at-fault traffic crashes. Also, we restricted the traffic crash data to only those crashes that occurred in the 11 SCP cities.
Table 22 provides the results from this benchmark. This pattern held for seven of the SCP locations. Crash data is for Muskegon rather than Muskegon Heights because the latter does not exist in the traffic crash database. Gray highlighting indicates that the percentage of stops involving African- American drivers is consistent with what would be expected based on their representation in in not-at-fault crashes.
The crash benchmark analyses provide important insight concerning traffic stop racial disparity within the SCP locations. However, in the same manner as discussed earlier, the benchmarks are. The findings remained unchanged and, therefore, were not sensitive to the types of crashes included in the benchmark analysis.
This may artificially deflate the percentage of African-American drivers represented in the crash data. It would be more ideal to benchmark the racial composition of traffic stops in SCP locations to the racial composition of traffic crashes that occurred in the specific patrol areas that troopers are assigned in those cities. Gray highlighting indicates that the percentage of stops involving African- American drivers is consistent with what would be expected based on their representation in in at-fault crashes.
Supplemental benchmark analysis without SCP location traffic stops To further test the robustness of the main findings described throughout the report, we excluded the SCP-related traffic stops from the analyses and reran several of the benchmarks from earlier in the report.
This removed 19, stops and left us with , stops for re-analysis. First, we reran the benchmark from Table 6 that compared the percentage of traffic stops involving African-American drivers to the percentage of African Americans in the population. While the percentage of African-American drivers involved in traffic stops changed slightly, the substantive.
African-American drivers were significantly more likely to be stopped across Michigan and within Districts 1, 2, 3, 5, 6, and 7 than we would have expected based on their representation in the respective populations a table of these findings is not provided in the report but is available upon request of the lead author.
Within District 8, African-American drivers were stopped at a rate that we would have expected based on their representation in the population. Again, while the percentage of African-American drivers involved in traffic stops changed slightly, all the substantive findings remained unchanged.
Given that daylight only predicted the odds of being an African-American driver in the main VOD analysis, we only estimated models predicting whether the driver was African American in these SCP-specific analysis. Table 24 presents the findings from this analysis. Model 1 is the logistic regression equation estimated using the traditional VOD analysis i. Again, this evidence suggests that African Americans were significantly more likely to be stopped during the day when it was, presumably, easier for troopers to see the race of the driver.
Model 2 in Table 24 presents the results from the VOD analysis restricted to those stops that occurred in the days before and after the switch to DST. Similar to the results discussed earlier, daylight no longer predicted the odds of the driver being African American while using this restriction.
Again, this finding provides caution when interpreting the main VOD results Model 1. Day of the week b Monday -. Intercept -, Pseudo R2. The earliest stops in the intertwilight period were coded 1 and the latest as 8. SCP stop outcome analyses Finally, we conducted the stop outcome analyses for only the 19, traffic stops that occurred in the SCP locations see Table Several different findings emerged in this analysis.
Most importantly, we see that African-American drivers were significantly more likely to receive a warning than their White. Intercept -. Note: Entries are unstandardized regression coefficients b , robust standard errors adjusted for clustering in the 8 Michigan counties that SCP stops took place in SE , and odds ratios OR.
This reduced the analytic sample to 19, traffic stops. The results from Model 3 in Table 25 search show that African-American drivers were no more likely than Whites to be searched.
This finding differs from the main analyses presented in Tables 18 and Lastly, in Model 4 of Table 25, African-American drivers were no more likely than White drivers to be arrested after the traffic stop in the SCP locations. Again, this finding differs from those presented in Tables 18 and Recommendations The final section of the report provides recommendations to MSP regarding traffic stop data tracking and reporting procedures.
The recommendations are in no particular order, but we attempted to group them into similar categories. Driver Race and Ethnicity Driver race and ethnicity are two of the most important data fields in a traffic stop database for external benchmarking analyses. Without accurate and complete information on driver race and ethnicity, benchmarking strategies are either flawed or impossible to complete. MSP prohibits troopers from asking drivers to self-report their race or ethnicity.
This creates a situation where there is room for error and inaccurate reporting. This is problematic in its own right but becomes especially challenging for benchmark analyses. Next, we discuss several issues we encountered with respect to the reporting of driver race and ethnicity within the traffic stop data. The eDaily data dictionary does not clearly provide a code book for these categories.
Rather, only letter designations are provided in the dictionary and within the data. Some of the letters are intuitive, but the research team had to communicate with MSP to verify what the letters represented. This will help external entities avoid mistakes when using the data. This is problematic for at least two reasons. First, race and ethnicity are distinct classifications.
People who identify as Hispanic ethnicity also will have a racial identity e. This would remove the burden from troopers, ensure accurate reporting of critical information, and allow for more precise benchmarking and other analyses in the future.
The research team encourages MSP leadership to communicate the importance of this issue to appropriate officials in the State of Michigan. Data Reporting Practices There were several data reporting practices that could be addressed to improve the overall quality of data collected by MSP.
A researcher could use this data in combination with information about the county where the stop occurred, but it is still limited.
With the current reporting practice, it is impossible to determine the exact location of a traffic stop. Such information can be valuable for racial and ethnic disparity analyses. We have several recommendations concerning the reporting of searches and other outcomes that occur during traffic stops. Current software allows troopers to provide additional information about searches, but it is not required to do so. Searches must be documented and accountability mechanisms for failure to do so must be in place.
Failure to properly track searches harms transparency with the community. Relatedly, the outcome of a search should be tracked within the MSP traffic stop database. Currently, there is not a field that indicates whether contraband was discovered as part of the search. Researchers and analysts would need to read the traffic stop narratives to determine whether contraband was found. Analyses have revealed that significant disparities in hit rates are found even in situations where benchmarking disparities are not observed COPS, The following categories could be included: o Seizure resulting from search?
Similarly, the reason for an arrest should be tracked. The current data indicates whether an arrest occurred but does not specify the reason for the arrest.
Data Documentation We encountered several data documentation issues that are worth noting. The eDaily data dictionary provides descriptions of the assignments but they are largely unhelpful. The purpose of a data dictionary is to provide lay people especially external entities that may be inspecting the data with a firm understanding of what specific codes mean.
Providing better descriptions in the data dictionary will help prevent frequent questions from external entities and ensure MSP employees correctly understand the coding. Accounting for the reason for a traffic stop is important within benchmark and post-stop outcome analyses.
The MSP traffic stop data provide a lengthy list of reasons why a trooper could have initiated a traffic stop. However, there are several hundred reasons in this list which makes recoding difficult and introduces the possibility of inconsistent analytic procedures as people conduct benchmark analyses and other analyses over time. This was done because the field allowed for a rough comparison of whether the stop was for a violation that endangered others or for a non- dangerous e.
This is useful in some respects, but it is insufficient for many of the questions benchmark researchers may have. This may be true in some situations e. For example, a broken taillight could be more accurately described as a non-moving violation. First, it would allow for more fine-grained analyses.
This is especially important in benchmarking because the purpose of such methodologies is to determine if racial or ethnic disparities can be accounted for by legitimate factors.
It is possible that accounting for more reasons for stops would help explain some of the observed disparities. Create Alert Alert. Share This Paper. Figures and Tables from this paper. View 9 excerpts, references methods. Benchmarking big data systems: A survey. BigDataBench: A big data benchmark suite from internet services.
View 2 excerpts, references background. Big Data Benchmark Compendium. View 1 excerpt, references background. Data-Centric Benchmarking.
Table Union Search on Open Data. VLDB Endow.
0コメント