Part V) Recommendations¶
We have built a model and analyzed which variables seem to be the most important for predicting graduation rates. Now what can we recommend based off of these insights?
1) Household Stability and Economic Security
We've mentioned this a few times, but we're going to hammer this home again: household stability and economic security are not only important factors, but they are the most important factors. That being said, what can be done to improve household stability and economic security?
1) Community development corporations (CDCs) are doing some great work in the community space to strengthen communities from a financial and social perspective.
2) Initiatives like the Low Income Housing Tax and the New Markets Tax Credit have brough vital services to low-income communities. Check out this article for more information.
3) Schools can step up and introduce programming that is targeted at students who are experiencing rapid or dramatic changes in their household. The goal would be to give them tools to help them cope, continue to allocate resources towards their education, and build a support network. (It might be the case that if support spending was allocated to programs like this, support spending would positively impact graduation rates.)
4) Schools can also step up by providing support to parents who are part of families going through dramatic transitions. This could involve helping parents build skills to deal with stress, remain nurturing towards their children during the transitions, and find other channels of support if needed.
5) State governments could also look into expanding conditional cash transfers for going to school. This would not only incentivize families to send and keep children in school, but also provide addional income that could increase a household's level of economic security.
2) School spending
1) We found that more instrucional spending tended to increase graduation rates, while other types tended to decrease it. Thus, all else being equal, it seems that schools should allocate more money towards instructional spending rather than support spending or administrative spending.
2) However, we also saw that our model predicts that graduation rates would not change if we increased spending. This suggests that instead of spending more money, we may want to focus on more qualitative factors, such as improving and updating curriculum, allowing for flexibility in course design, etc. We don't want to throw more money into a process when it won't generate results. But if we can create a better process, more money could translate into better higher graduation rates.
2) Our findings also might point to a deficiency in the support programs provided by schools, which suprisingly seemed to have a negative effect on graduation rates. Thus, another action item would be to redesign support programs, or at least figure out which ones work better, and focus attention on them.
3) Inclement weather
Not much research has been done to examine the relationship between weather and graduation rates. That being said, here are some ways schools can protect themselves against inclement weather:
1) Colder areas tend to have lower graduation rates than warmer areas. One way to design around this problem is to have alternate school schedules for colder areas. It might be better to have a longer break during winter months and a shorter break during summer months for these schools.
2) Schools in cold areas in particular should look to online education as a way to keep students engaged in classes despite inclement weather.
3) Schools in colder areas should also consider infrastructure upgrades. It might be that schools with more robust walking and school transportation infrastructure have lower absentee rates.
4) Graduation rate data
For this analysis, we needed to drop many of our observations because of imprecision in the dependent variable. There might be other ways to get around this obstacle, such as imputing imprecise variables based off of other characteristics in the dataset. However, in order to build better and more accurate predictive models, it would be best to have complete, accurate, and precise data.
We realize that there are privacy concerns with releasing graduation rates, but there are other ways to get around that. For example, we could mask the graduation rates with some sort of encryption code to prevent people from figuring out the district that that graduation rate is attributed to. The point is that there are other ways to alleviate privacy concerns.
Conclusion and further analyses¶
This analysis sheds some light on where states, cities, and schools should focus their attention on when trying to improve graduation rates. However, more work can be done to deepend our understanding of which factors are relevant. Some further analyses that would be fruitful are:
a) What kind of household instability matters the most? We used variables such as female-only householder households and umarried households, but those are really only proxies for household instability. Household instability is a qualitiative assessment determined by rapid and dramatic transitions and disruptions in a household. There are probably plenty of unmarried households that are relatively stable. Instead, we might want to look at more specific variables, such as divorces, the amount of time single parents invest in children, frequency of relationship turnover of a single parent, etc.
2) School spending data should get more granular. For example, while the student support spending variable includes spending on social services, it also includes other types of student support that were more adminstrative. It would be good to separate out spending on social services vs. spending on other services that fall under student support expenditures. That way, we can get more accurate estimates of the effects of, say, social services.
Alternatively, it would be good to run randomized controlled trials on specific programs run by schools and evaluate the effectiveness of those in order to understand what we should scale up and what we should cut back.
3) With regard to weather data, we should look at the precise causal mechanism between colder weather and lower graduation rates. Is it because of school cancellations? Or lower attendance rates? Or some other reason? Once we understand the causal mechanism, we can better tailor policy recommendations.
The US aims to reach a 90% graduation rate nationwide by 2020. It's doable, but more work needs to be done. We hope this analysis can help us reach that goal.