Statistics (1 Viewer)

[jm1234567890]

Where to you get the average cohort agregate and the standard deviation?

 

[Lazarus]

They aren't published - though you could try writing to the Technical Committee on Scaling.

I haven't yet got around to doing so myself.

 

[jm1234567890]

i have them for 95/96/97....

you mean this year's isn't published?
could you post 2001/2002 ones

 

[Lazarus]

They haven't been published since the introduction of the UAI in 1998.

The ones you have would be for the old TER.

 

[jm1234567890]

hmm....

TER and UAI are similar.

without those figures how do you calculate UAI then....
In the formula i got, you need them to find UAI.

I can't really see a way do get around it.

EDIT: i can calculate the agregate, but i'm not sure about agregate --> UAI

 

[Lazarus]

Similar, but not the same.

I'm not even sure how you'd use the mean and SD to calculate TERs... perhaps with a normal distribution table, but I don't really know how accurate that would end up being. An interesting thought though... might look into it.

You can't calculate UAIs using the mean and SD of the aggregates, however, because the aggregates only correspond to the HSC cohort, whereas the UAI ranks students against their SC cohort.

I don't really want to just give you my method for converting aggregates to UAIs, because that sort of defeats the purpose of you creating your own calculation program. (Or it at least removes the reason I had for wanting you to do so.) But I believe there is enough info in the public domain to devise at least a handful of different methods.

 

[jm1234567890]

Damn the UAI Vs Agregate cure isn't a normal one.

I'm going to try and use known agregate/UAI values and try to fit it to a curve.

Since the values do form a smooth curve (from the technical report).

The problem is the type of curve to use... this is going to take a while

 

[jm1234567890]

My, scaling algoritms seems to be a bit off too.

I think i check that first.

Unless your's is wrong

BTW, i'm reftering to UAIseeker2003

EDIT: also, i have read through all possible documents that i have found

EDIT: "A Bit" meaning 0.1 - 1 off

EDIT: laz seeker seems to scale the same as UAIseeker

 

[Lazarus]

Originally posted by jm1234567890
My, scaling algoritms seems to be a bit off too.
...
EDIT: laz seeker seems to scale the same as UAIseeker

They both use the same scaling algorithm. It merely approximates the scaled mark from the aligned mark; there would be more than one way of making this approximation. Your method might be just as valid.

 

[jm1234567890]

nah, mine is wrong.

mine simply uses a linear transformation, the real values don't seem to follow that.

yours seems to use the real values and fit a curve to it.
However, i'm not too sure your vaules for <25percentile are right. They seem to drop off rapidly

For the agregate VS uai, an arctan curve seems to fit it well. i'm going to get some 2002 agregates and try it out.

 

[Lazarus]

Originally posted by jm1234567890
However, i'm not two sure your vaules for <25percentile are right. They seem to drop off rapidly

You're probably right there... I think we assumed that the bottom mark in any particular course would be 0. I guess it could be set higher, but it's really quite arbitrary.

 

[jm1234567890]

i reckon i have the maths for it, now i need to get that damn pdf into something java can read