MAKE AN APPROPRIATE FORMAL STATISTICAL INFERENCE

MAKE AN APPROPRIATE FORMAL STATISTICAL INFERENCE

To overcome this problem we use a technique called bootstrapping. Bootstrapping involves taking a small sample representative of a population and then using this sample as a mimic of the population by taking even smaller samples (with replacement) from within that sample. After each resample the difference between the medians of each group is recorded. We can then build up a picture of the variability of these differences. A dot plot shows the distribution of these median differences. By taking (approximately) the middle 90 to 95% of these medians we can create a confidence interval that we believe will capture the population parameter that we are attempting to estimate fairly accurately. In this case we wish to estimate the difference between the median birth weights of VLBW babies that survive and VLBW babies that do not survive.

The following bootstrap confidence interval is based on 1000 resamples.

By using this bootstrapping method we can be reasonably sure that the median birth weight of VLBW babies that survive is somewhere between 257.5 - 410g heavier than the birth weight of VLBW babies that do not survive.

As this interval does not contain zero we can be fairly confident that on average VLBW babies that survive are heavier than VLBW babies that do not survive.