Result and Discussion
Result
Finally, we arrive at our results using the transformed data with the difference in differences technique.
We first compared on state level. By looking at the delta, it indicates there is only 0.22 percent difference in percentage of DUI cases before and after RML in California. The t test indicates that small amount of difference is not statistically significant. This is actually the opposite from what most of the news tells us, which RML would increase the number of collision cases and therefore make the society more dangerous.
| Coefficient | Standard error | t | P>|t| | [0.025 | 0.975] | |
|---|---|---|---|---|---|---|
| Constant | 0.0026 | 0.001 | 2.465 | 0.016 | 0.000 | 0.005 |
| Post | -0.0044 | 0.001 | -3.227 | 0.002 | -0.007 | -0.002 |
| Treat | -0.0013 | 0.001 | -0.891 | 0.375 | -0.004 | 0.002 |
| Interaction | 0.0022 | 0.002 | 1.166 | 0.247 | -0.002 | 0.006 |
We also tested out at the rural county level by filtering the states’ data to rural areas where population is low. We hope that these low population areas would share some similar patterns or behaviours so that these traits could reduce the influence of other confounding variables. The result, again, shows a negligible difference before and after RML. Therefore, we could not conclude our hypothesis, which RML would bring an increase in percentage of DUI cases.
| Coefficient | Standard error | t | P>|t| | [0.025 | 0.975] | |
|---|---|---|---|---|---|---|
| Constant | 0.0011 | 0.002 | 0.749 | 0.456 | -0.002 | 0.004 |
| Post | -0.0019 | 0.002 | -0.981 | 0.329 | -0.006 | 0.002 |
| Treat | -0.0016 | 0.002 | -0.774 | 0.459 | -0.006 | 0.003 |
| Interaction | 0.0027 | 0.003 | 0.974 | 0.333 | -0.003 | 0.008 |
Conclusion
Overall, we did not have enough information to conclude that RML has a significant impact on California's percentage of DUI cases among all collisions. This finding is on both state level and rural county level. Through this project, although we did not get any significant results, we realized sometimes news could be misleading. They need to be catchy in order to bring in more audience rather than plainly stating the fact.
Discussion
One of the main limitations for this project is the control group. We are not able to find a good control state among all states that could hold the common trend assumption perfectly. We believe as more states start to realize the importance of the use of data, there would be more data source for us to find a better control group which would ultimately provide the robust result.
In addition, this project is just a beginning on finding out the impact of legalizing marijuana. We would further expand this project on other states that legalized marijuana. In that way, we could get a comprehensive understanding on whether RML would bring a significant impact or not across United States. If there is significant result on certain states then we could dig into that more and find out what's the cause of that sigificance.