I'm saying that the conclusion drawn doesn't follow from the experiment.

As I said, to be sure that skin tone was the key factor, and not simply the only possible difference to distinguish against, they need more colors (even unrealistic ones), and varying traits that can influence the decision as well.

With the test currently set up as it is, you force the decision to be about the color of the doll, given that it's the only difference. So really the test could only go one of three ways: white=evil, black=evil or a roughly 50/50 split. As it so happens, we get roughly a 50/50 split, with a bit of a lean on black=evil.

That doesn't say or prove anything at all.

Is there still plenty of racism to go around? Sure. Is this racism being taught to kids? Sure. Does this experiment demonstrate racism? Nope. The sample size is also pathetically low. Only 20 kids? You could easily round up 20 kids, prepare them by giving them answers, and rewarding those answers after the experiment. Then only film the responses. You'd get the exact same response.

As I said, the phrasing is awful (implying you must pick one), the set up is awful (having the only difference be skin color, thus making it the distinguishing choice), the filming is awful (cutting away from kids that don't give the proper response), and so on. So many problems with the experiment itself to be able to draw anything from it.