Here’s a video of the experiment:

The researchers weren’t certain that the toddlers were picking the block based on probability, rather than frequency. In the initial experiment, each block was placed on the platform six times—one of them dispensing three marbles, the other dispensing two.

In order to discount that the toddlers were choosing based on frequency, the researchers ran the experiment again, this time dispensing the same number of marbles for each block (while maintaining the same probabilities, one-third and two-thirds). This time, 22 of the 32 toddlers still picked the more successful block.