Paper cuts: Do ambient light levels influence criminal activities?

Written by Brian Tarran on . Posted in Social Sciences

People often feel less safe when walking in the dark, for understandable reasons. It’s harder to be fully aware of your surroundings, for one, which causes many of us to fear a criminal might be lurking in wait. But are we right to be concerned? Do light levels influence criminal activity? That’s a question asked recently by Jennifer L. Doleac of the University of Virginia, and Nicholas J. Sanders of Cornell University.

The two economists were inspired to investigate this issue by news stories they had read during the recession of the late-2000s. In a bid to cut costs, governments were shutting off streetlights. But Doleac and Sanders wanted to know whether this was a smart move, economically speaking. “Should policy-makers have expected crime to change as a result? How would this affect the cost-effectiveness of their budget-cutting strategy? Those are tough questions to answer,” they wrote in an email to Significance. “A lot of factors influence how well-lit a particular area might be, such as local funding, demand from residents, and local crime trends. So understanding the effect of light on criminal activity requires more than a simple comparison of areas with and without good lighting.”

Fortunately, the US government had provided them a natural experiment to work with. In 2007, daylight savings time was extended by four weeks: clocks rolled forward three weeks earlier in spring, and rolled back one week later in fall. By looking at crime data for that three-week period in spring1, Doleac and Sanders were able to compare the number of recorded incidents in hours that were darker pre-2007 against hours that were lighter post-2007.

They found that “in the weeks after the start of DST, the overall daily robbery rate (in terms of robberies per 100,000 population) decreased by around 7%”. Curious to know more, we asked (and they answered) the following questions:

Prior to your research, what evidence did we have about how light affects criminal behaviour/activities?
The data on crime across the day show a correlation between light and street crime – crime rates are higher at night than they are during the day. However, there are clearly other differences between the day and night, and even gradual change in sunrise and sunset times over the year correlate with other seasonal trends. Simple correlations don’t tell us that less light means more crime.

Much of the previous literature focused on streetlights, but effects were typically not well-identified. For instance, researchers looked at the difference in crime before and after neighborhoods added streetlights. They usually found crime fell at night, but also found crime fell during the day, when street lighting shouldn’t play a role. This suggests it wasn’t the lighting that made a difference, but that streetlights are a signal of investment in a neighborhood, or the timing of installation correlated with underlying crime trends.

Further work compared neighborhoods with new streetlights to “matched” control neighborhoods. That evidence was quasi-experimental but one still wonders why one neighborhood got streetlights when the other didn't. Overall, it was very difficult to find something akin to random variation in lighting that would really nail down the effect.

The only earlier research on daylight was a report following a year-round DST extension in 1974, comparing crime in Washington, DC, from one year to the next. It found violent crime was 10-13% lower during the year with extended DST than in years without. This provided suggestive evidence about the crime-reducing effects of evening daylight, but of course many things change over the course of years so the study was far from conclusive.

You started with the hypothesis that the increase in daylight brought about by DST would reduce crimes at specific times of the day. What crimes did you focus on and why? Also, how did you test this hypothesis?
We used data from the National Incident-Based Reporting System (NIBRS) because it contained detailed information on the time that crimes occurred. By using specific hour of the day (rather than just the date) to classify crime timing, we were able to investigate how shifts in daylight changed crime rates during different hours of the day. If the mechanism was really about changes in ambient light, we expected effects primarily during the hours of sunrise and sunset (which were affected by the time change). We didn't expect anything different before and after the start of DST for hours of the day that were always light (e.g., 1 pm) or always dark (e.g., 11 pm).

Similar logic led us to focus on robbery. It is a face-to-face crime where the danger of observation and recognition can play a larger roll in the criminal decision, and it is a common street crime. For example, the data on robberies include muggings. It is also a crime where the hour of occurrence and the hour of reporting likely line up. Other crimes, while important and interesting to study, are less well suited for our particular source of variation and identification. Fraud, for example, includes less concern of recognition for those committing the crime, and therefore isn't subject to changes in darkness. Early in our analysis we used data on fraud and swindling as "logic checks" to make sure we didn't see a change around DST where our model dictates we shouldn't. Breaking and entering or auto theft likely correlate with darkness, but time of reporting will vary based on when the victim discovered the crime (when people notice things missing or wake up in the morning and find their car damaged) rather than when the crime occurred. That makes a focus on hourly effects less applicable.

What did the analysis reveal?
When we looked at hourly crime rates, we found the only times with statistically and economically significant changes in crime rates were right around sunset. In the weeks after the start of DST, the overall daily robbery rate (in terms of robberies per 100,000 population) decreased by around 7%. Going along with our hypothesis about the mechanism being shifts in ambient light, that change is a product of a 27% decrease in robberies during the hours right around sunset.

You give an estimate for the economic cost saving generated by DST. How was this calculated?
Criminal activity has many costs to society, and other researchers have thought carefully about how to measure those costs. The estimates we use, from work by McCollister, French, and Fang (2010), include tangible costs, such as medical expenses, lost wages, and costs to the legal system, as well as intangible costs such as victims’ pain-and-suffering. Those all add up to about $42,310 per robbery. By reducing the number of robberies, we avoid those social costs. Since our paper considers a 3-week extension of DST in 2007, we multiplied our estimated reduction in robberies (0.215 per million residents per day) by the US population (roughly 310 million) by 21 days by $42,310. That comes to a $59 million social cost savings for those three weeks.

We also find suggestive evidence that DST reduces rape, where each reported rape has a social cost of $240,776; if we include that effect, the annual social cost savings total $246 million over those three weeks.

Your headline finding on robbery reduction is significant at the 10% level, which is above the standard level of statistical significance (5%). What gives you confidence in this finding?
As economists and researchers, we worry about both the statistical and economic significance of findings. An effect might be very precisely estimated but so small as to be of little policy consequence, or might be large enough to warrant attention even if the estimate is noisy. There’s no a priori reason to believe a p-value of 0.05 is the end-all cutoff for numbers about which we care when building policy. But we do want to be careful about separating meaningful outcomes from statistical noise.

There are a number of additional factors that make us more confident that what we find is not just statistical noise. First, our results by hour, where we expect the greatest precision, are statistically significant at 5%. Second, our results on the probability of any crime occurring on a given day are statistically significant at 5%, which supports the hypothesis of a net increase in crime when evening hours are darker. Finally, using an alternate difference-in-difference estimation design, we get approximately the same values as our regression discontinuity model with results that are statistically significant at 1%.

What are the policy implications of your research?
The main policy implication is that ambient light plays a role in crime rates, and shifting daylight from the morning to early evening hours appears to reduce street crime. We find criminal activity isn’t simply shifted to the (dark) morning hours when DST is in effect; this is what makes DST good at reducing crime rates overall, and suggests that extending DST could be a simple way to reduce street crime.

But when considering whether to extend DST further, we need to consider the other potential costs and benefits. For instance, other researchers found increased heart attacks and car accidents around DST, and there are substantial costs to adjusting schedules. But those costs are due to the transition from DST to Standard Time and back – our internal clocks are thrown off by the time change, and that causes stress and confusion. There might also be costs associated with a permanent change, due to things like people driving in traffic in the dark, or children waiting for the bus or walking to school before sunrise. These should be studied further. Our read of the current literature is that most of the costs associated with DST are due to the transition, and could be avoided if we made DST permanent.

Regarding streetlights, our research suggests that areas should think twice about reducing lighting, even when budgets are tight. Cutting down on a power bill might look like a good idea on paper, but the harder-to-observe social costs could easily outweigh the savings.

What are the limitations of your findings? For instance, the data set used seems to exclude large, well-lit cities, which might not see the same DST effect on robbery rates.
External validity is always a concern when using unique situations to identify general effects. You make a great point that the data we have, while the best available and very rich, don’t cover as many locations as we would like. It is possible the areas NIBRS covers are not representative of other cities not in the dataset. To generalize the results, we must assume that daylight affects criminal behavior similarly in those cities.

Similarly, if one wants to make broad statements about lighting policy, one has to be comfortable saying that sunlight and streetlights have similar effects on crime, which may not be true. There’s also the interesting question of general equilibrium effects. Our research focused on just the weeks around the beginning of DST, where variation is the greatest. But can we comfortably say this extends into the middle of the summer? If we moved to a year-round daylight saving time model, would people eventually adjust their behavior? Ultimately, these are not questions we can answer with certainty, given our data and empirical strategy.

Footnote

  1. The fall period was excluded from Doleac and Sanders’ analysis because they were concerned that Halloween might act as a confounder.

What is 'Paper cuts'?

Interviews, reviews and reflections on statistics journal papers, both new and old. To submit a paper for consideration, or to write your own review of a study (as long as it's not a review of your own work), please email This email address is being protected from spambots. You need JavaScript enabled to view it., with the subject line 'Paper cuts'.

 

Subscribe

Significance magazine is available by subscription or as a member benefit in joining the American Statistical Association, or the Royal Statistical Society.

Please visit the website of the American Statistical Association or the Royal Statistical Society for information on becoming a member or to renew your membership.