1: \begin{abstract}

2: We propose a nonparametric statistical test for goodness-of-fit:

3: given a set of samples, the test determines how likely it is that

4: these were generated from a target density function. The measure of

5: goodness-of-fit is a divergence constructed via Stein's method using

6: functions from a reproducing kernel Hilbert space. Our test statistic

7: is based on an empirical estimate of this divergence, taking the form

8: of a V-statistic in terms of the gradients of the log target density

9:  and of the kernel. We derive a statistical test, both for i.i.d.

10: and non-i.i.d. samples, where we estimate the null distribution quantiles

11: using a wild bootstrap procedure. We apply our test to quantifying

12: convergence of approximate Markov chain Monte Carlo methods, statistical

13: model criticism, and evaluating quality of fit vs model complexity

14: in nonparametric density estimation.

15: \end{abstract}

16: