Do dogs prefer petting or verbal praise?

Researchers at the University of Florida collected data to try to answer this question.^[1] They measured the amount of time a dog spent interacting with its owner during a five-minute period while the owner was offering petting for a sample of seven adult dogs. They also measured the amount of time a dog spent interacting with its owner during a five-minute period while the owner was offering verbal praise for a sample of seven adult dogs.

If a two-sample [latex]t[/latex] confidence interval is not appropriate, what can we do instead? It probably won’t surprise you that we could use a bootstrap confidence interval!

When we have two samples and want to estimate a difference in population means, we focus on the difference in the two sample means. If we had chosen different samples of the same size, we would have seen different sample means and a different value for the difference in sample means. What differences in the sample means could have arisen?

Bootstrapping constructs more possible differences in sample means by generating a bootstrap sample from each of the original samples and calculating the difference in the bootstrap sample means. If this process is repeated a large number of times to form a bootstrap distribution, we can use that distribution to construct a [latex]95\%[/latex] bootstrap percentile confidence interval by identifying the [latex]2.5\%[/latex] and the [latex]97.5\%[/latex] percentiles from the bootstrap sampling distribution.