Really cool article! I find your empirical design super compelling and love how you motivate the question and introduce the data and results.
I wonder what Graph 1 would look like using a box plot. It would also be interesting to test whether other teams experienced the same slump with win/loss data for all teams. You could regress win percentage (Y) on the share of night games (X), a dummy indicator for late-season months (Z), and an interaction between the share of night games and the dummy for late-season months:
Y = a + b*X + c*Z + d*X*Z + e ----> reg Y X i.Z X##i.Z
The coefficient d would capture the marginal association between night games and win percentages in late-season months. An even simpler design would be a regression of the difference in win percentages between late- and early-season months on the share of night games.
Lastly, I wonder how omitted variables might threaten your causal interpretation. Is it possible that baseball teams in cold-weather climates are both less likely to have night games (because they do not need to avoid scorching daytime temperatures) and more likely to experience late-season slumps (because colder weather impacts performance as well)?
Again, great article and I look forward to seeing what else you publish!
Thank you! Because they’re playing against each other, it’s impossible for the league as a whole to slump; somebody would have to do better. So that is why I think only testing if our team of interest slumped is relevant.
The causal interpretation is indeed fragile. I hope my evidence is quite suggestive of it being a real effect, however.
Really cool article! I find your empirical design super compelling and love how you motivate the question and introduce the data and results.
I wonder what Graph 1 would look like using a box plot. It would also be interesting to test whether other teams experienced the same slump with win/loss data for all teams. You could regress win percentage (Y) on the share of night games (X), a dummy indicator for late-season months (Z), and an interaction between the share of night games and the dummy for late-season months:
Y = a + b*X + c*Z + d*X*Z + e ----> reg Y X i.Z X##i.Z
The coefficient d would capture the marginal association between night games and win percentages in late-season months. An even simpler design would be a regression of the difference in win percentages between late- and early-season months on the share of night games.
Lastly, I wonder how omitted variables might threaten your causal interpretation. Is it possible that baseball teams in cold-weather climates are both less likely to have night games (because they do not need to avoid scorching daytime temperatures) and more likely to experience late-season slumps (because colder weather impacts performance as well)?
Again, great article and I look forward to seeing what else you publish!
Thank you! Because they’re playing against each other, it’s impossible for the league as a whole to slump; somebody would have to do better. So that is why I think only testing if our team of interest slumped is relevant.
The causal interpretation is indeed fragile. I hope my evidence is quite suggestive of it being a real effect, however.