I have one state’s school reopening data and have been running some ML algorithms with cross validation to predict school reopenings. The SafeGraph features heavily in these algorithms. I will hopefully be getting some national data soon, and am wondering if anyone knows of any data that could help me better tune my reopening prediction models. Thanks!
Hi @Trevor_Gratz_University_of_Washington, could you let me know some more details on the data you are looking for?
Hi @Jack_Lindsay_Kraken1, we are looking for any national panel data (at the month level or finer) that could help us discriminate between open and closed schools districts.
For example, we are using 2 years of SafeGraph data to compare 2019-2020 school year months to 2018-19 school year months. We used this data to create a ratio of device traffic data, Month_m_y / Month_m_y-1, where m indexes a month and y a year. This data along with Common Core Education data (more or less demographic data on students) and month fixed effects does a reasonable job at classifying whether or not a district is open in a given month (auc ~.87).
I was hoping someone might have come across another dataset or feature that could help us better predict when school districts had reopened. In particular, the SafeGraph data is noisy for smaller more rural districts and our predictions are worse in these areas. Ideally this new data would run through Feb.
Any insight would be very helpful, thanks!
Hi @Trevor_Gratz_University_of_Washington, sorry for the delay here - I was hoping someone would have a better answer than I do. Do you think you could gain any insight by checking out some recent research done HERE - THIS THREAD may also help
Apart from this, I am not sure. I wish I could be of more help
Hi @Jack_Lindsay_Kraken1, thanks for pointing this out! I am going to be attending a seminar on that recent research tomorrow, so I am looking forward to learning what they did. Thanks!