how could this be done using numpy/pandas in jupyter lab?

4. In competitive marathon, it's generally better to have a negative split (although very hard to do), which is to run the second half of the race faster than the first half. What percent of the runners in our data had a negative split? Round your answers to 3 decimal places