Here is the data we gathered:
Note: We initially had 102 candies in our sample. I told the students that it would be easier to figure out the expected counts if we had 100 M&M's. They suggested removing two candies from our collection "at random" with the help of random.org.
So, where was the difference? Which colors were the largest contributors to the test statistic. When we ran the test on the calculator, it created a list called "CNTRB", which contains the individual addends that were used to create the test statistic. The largest value was about 11.6 for yellow and the second largest contributor was 4.4 for the orange candies. If you look at the observed and expected counts, this makes sense.
Also, you would think that the colors would be made in the same proportions at both plants, but they aren't. They are close, but they are not exactly the same. The screenshot below shows the proportions of each color from each plant. I am not sure when the percentages changed, but I do know that in 2008 the percentages were different than what is listed here.