Is a kernel density estimation a good approach for small samples?

If you have a relatively small sample of data points that has more than one mode and you want to estimate the distribution of the population it came from as well make inferences, is kernel density estimation the way to go?


Less is more. Stay pure. Stay poor.
If it is multimodal, perhaps it is still a mixture of variables, and you have not defined all components of the data generating process. Kernel densities are great at getting distribution estimations. Keep in mind that a small sample may not be the best thing to use to understand the true underlying population distribution. Simulation studies can show the diversity in sample distributions you can see in small samples randomly pull from the super population and many times our samples are not even random pulls compound this issue of working backwards.