Because you're afraid you'll lose. Good call. If only you weren't to chicken to admit it...
You have nothing.
If you did you would have shown me something more than a horribly under powered study by now.
You can't produce one thing to even suggest you are right.
Give me the slightest hint you are right.
I can produce normal curve after normal curve on all kinds of written tests and you know it.
The
raw scores rarely if ever have anything near perfectly normal curves. The curves you see are the product of often quite intricate normalization algorithms, on top of using a question set specifically designed to produce as closely as possible a normal distribution. If you want to learn more about the kind of algorithms typically employed, go here:
http://www.ets.org/Media/Research/pdf/Angoff.Scales.Norms.Equiv.Scores.pdf If you want me to spoonfeed this information to you, you're henceforth going to have to pay.
Using tricks of similar complexity, one could just as easily force a bimodal distribution. Indeed, the raw scores sometimes do show a bimodal distribution. You don't have to take it from me, take it from Mr John Raven of the Raven test fame: " First, as was evident from Fig. 2 the within-age score distributions for the RPM (and, according to a personal communication from Robert Thorndike, the subscales of the Stanford–Binet test) are generally not Gaussian and are, indeed, often bimodal." -
http://eyeonsociety.co.uk/resources/RPMChangeAndStability.pdf
Even when the raw scores aren't bimodal, a simple reweighting of the per-question results can always force a bimodal result (unless the questions are all of equal difficulty and there is no inter-correlation of results): Just pick a handful of questions almost everyone was able to answer, and multiply the score of each of them by 2. then pick a subset of intercorrelated difficult questions where most people either answered all or none (because they're of similar difficulty and question the same kind of reasoning), and multiply their weighting by 2. Multiply all other questions' weighting by 0. Voilá, a peak at 10 and a secondary peak at 20. And if
that doesn't work (it wil), pick
one question towards the difficult end of the scale and multiply its weighting by 10: again: a peak at 10 for the people who solved all of the easy questions but not the difficult one, and a secondary peak at 20 for the ones who solved all of the easy ones plus this one difficult question.