by Bill Betzen
This past week the Dallas Independent School District released current enrollment numbers which indicate very positive progress in lowering the dropout rate. Using these enrollment numbers it was determined that the Cumulative Promotion Index has gone from 46.6% last year to 54.0% this year! This 7.4 percentage point gain is wonderful! Current enrollment numbers indicate it will continue to go up next year as well.
Other good news is that the improvements were spread all across Dallas, and were especially positive in North Dallas where the 6 high schools went from an average 9th to 12th grade promotion rate of 39.6% in 2008-2009 to 53.6% for 2009-2010! The graph below shows this years progress compared with the past 11 years. DISD well on the road to setting new records with the 2010 graduation rate.
[Ed Note: Bill, I'm a little skeptical of this data. I'll clarify below...]
(Please pardon me, folks, while I geek out with Bill--and any other statisticians who want to get into the nitty gritty)
Bill, I'm going to exclude the latest year from the "CPI" on your statistics page (http://www.studentmotivation.org/DallasISD.htm) and also assume that your figures are based on population means as opposed to sample means. So n=12, and you're roughly normally distributed (very slight negative skew -0.10084) around a mean of 44 with a standard deviation of 4.1888 (but let's go ahead and lose a degree of freedom and take s=4.3751) thus making your standard error 1.263. So a .95 ci of your sample data is 41.52 to 46.48 (sorry, everybody, I warned you this would get geeky!)
OK so far?
So let's bring your 54% figure back into the picture and assume the null hypothesis to be H054. A simple one sample Z-test statistic is -7.918 with a P-value of almost 1.
Obviously we don't have a sample size > 30 so central limit doesn't apply. Even so, it's highly unlikely you'd see this observation unless you had an obvious explanation for it (elimination of a high-dropout group, for example). What's more likely is that DISD changed it's statistical model.
Bottom line: statistically this is extremely unlikely and we'd be right coming to this conclusion 99.9999999% of the time.
I'd be very interested in the raw data--but I strongly suspect we'll be missing some key pieces needed to validate this claim.