Very cool… I havent run my models yet, but does date recorded have a strong correlation with result? that could be a serious methodological issue if it does…
Yes, there is a slight kuznets curve relationship between date_recorded to the percentage of functional/non functional pumps. It does make intuitively sense also, older pumps break more often. What doesn’t make logically sense is that most of the oldest recorded pumps are still functional, but there are probably other variables beside date_recorded at play.
Thanks for the reply, there is certainly something interesting there! I should have time to run some models within the next two weeks, I would love some feedback once I post them.