Don’t larger images usually make it easier to hide a secret?
Yes, that’s usually true. Because you have more places to embed the bits, and you can use more LSBs producing usually a less noisy image.
Then why does the quality seem to go down in those cases?
The last is true if the host image is larger, but in the post, the host image remains always at the same size.
In the results, I’m increasing the size of the secret images thus increasing the amount of data to encode, while the quality of the resultant stego image decrease.
Ah, now I understand. Thanks for clearing that up!