Model

\[ \begin{aligned} &logit(P(response\space time > 8 min))\\ &= \beta_{0} + \beta_{1}I(season = Fall) + \beta_{2}I(season = Summer) + \beta_{3}I(season = Winter) \\ & + \beta_{4}I(hour\space of \space day = afternoon) + \beta_{5}I(hour\space of \space day = dawn) + \beta_{6}I(hour\space of \space day = morning) \\ & + \beta_{7}I(snow = 50mm+) + \beta_{8}I(snow = 50mm-) + \beta_{9}I(prcp = 25mm+) + \beta_{10}I(prcp = 25mm-) \end{aligned} \]

Result

Snow As expected from EDA, compared to no snow conditions, odds of over 8min response time is increased 24% by 0~50mm snow and 47% by over 50mm snow.

Season Compared to Spring, Summer does not affect odds of over 8min response time but odds is increased in Fall by 6% and in Winter by 7%.

Rain Statistically significant result was not obtained from rain variable and it matches the result of EDA.

Hour of the day In reference to night, odds of over 8min response time is increased at dawn by 26%, in the morning by 37% and in the afternoon by 35%.

Subset analysis

We fitted regression model by neighborhood and here is our findings:

  • On Upper West Side, high precipitation is associated with lower odds of response times greater than 8 minutes.
  • Street closure is associated with the outcome but not statistically significant. Washington Heights, on the other hand, saw afternoon and dawn time, compared to the night time as reference, as contributing factors to the response time greater than 8 minutes with ORs 1.6 and 1.8, respectively. Again street closure is not predictive of the outcome.
  • The null result from street closure could be due to either linking street closure to the zip code in this case is too broad or street closure did not really play a role in our response time.
  • The model with neighborhood as predictor variable tells us that Washington Heights had 1.05 times the odds of response time greater than 8 minutes compared to the Upper West Side. However the observed effect is not statistically significant. At least we can be rest assured whether where we live, we still receive equal treatment from the emergency dispatcher.

Final conclusion

From our exploratory and logistic regression analysis, we found out that in snowy conditions, the odds of over 8min response increases while rainy condition does not effect the outcome in New York City in 2017. Along with this result, the odds of over 8min response increased in winter and it may be linked to snow factor. From this analysis, we can recommend a plan which assigns more ambulances in snowy conditions and during early hours.

While comparing Upper West Side and Washington & Inwood Heights, the odds of getting different response time wasn’t statistically significant. Hence, we can be rest assured that where we live, we will receive equal treatment from the emergency dispatcher.

Further investigation can be done using traffic as a variable. This can also be specifically used to determine if holiday events such as Thanksgiving or Hawlloween Parades affects the response time.