Re: Predicting Student Test Scores

On Jul 28, 5:26?pm, D Herring <dherr...@xxxxxxxxxxxxxxxxxxx> wrote:
cynthia.lla...@xxxxxxxxx wrote:
What would be the best formula for predicting student test scores (if
there is one) based on past scores. For example lets say the
following students took six math tests and their scores were: (most
recent score listed first)

...

Which student would be most likely to score the best on the next math
test taking nothing else into consideration but the previous math
scores...is this possible to predict and how? Also, what if I wanted
to put more emphasis on the most recent two scores, how can I
compute....thank you in advance.

Umm... maybe a weighted least squares linear fit? Quadratic fit?http://www.itl.nist.gov/div898/handbook/pmd/section1/pmd143.htm

Most spread*** programs can "add trend lines" to plots of data;
these are generally least-square polynomial fits.

an RLS filter?http://en.wikipedia.org/wiki/Recursive_least_squares_filter

The simplest would be a weighted mean:
> Student A - 68 - 86 - 74 - 80 - 83 - 85
mean = (1.5*(68+86) + 1.0*(74+80+83+85)) / (2*1.5 + 4) = (231+322)/7 =
79 <- predicted next score

Good predictions require good models of behavior. People are hard to
model.

- Daniel

I think the simplest is good since I am a math novice but thank you
for the links....very interesting! One more question, what if I had to
compare a unequal amount of scores, lets say student B had only 5
tests..81 - 74 - 70 - 80 - 83 and we'll keep student A the same.. is
the following correct?
1.5*(81+74) + 1.0*(70+80+83)) / (2*1.5 + 3) = (232.5+233)/6 =
77.58 <- predicted next score

Why the "2" in (2*1.5 + 4)in your above answer?

thanks again


.