Instead of Providing Yearly B6's for ATAR Estimates, Why Don't We Also Look at Historic Averages [Suggestion] (1 Viewer)

C

carrotsss

New Member
Joined
May 7, 2022
Messages
4,691
Gender
Male
HSC
2023
The trouble with using a method like this with an average is that mean ≠ median, and for the vast majority of schools and subjects, marks can exhibit a positive or negative skew rather than perfectly symmetrical data. I found this myself while working on the raw atar calc, when I initially included the average aligned/scaled mark conversions in my modelling it was significantly different to the percentiles, and once I removed them it made quite a significant difference to accuracy - and that’s on the scale of an entire subject cohort, singular schools are significantly more prone to outliers. Band 6 counts are far from a perfect method either (and in an ideal world everyone would provide multiple years of data as that would improve accuracy by a fair bit), but the way they work off a specific percentile is a significant advantage.

A better data point I’d like to see more people include is band 5 statistics, which I’ve found many schools include in their annual reports, as they would aid significantly in ATAR estimates where people have ranks in that range.
 
M

MaccaPaccca

Member
Joined
Feb 11, 2023
Messages
25
Gender
Female
HSC
2023
carrotsss said:
The trouble with using a method like this with an average is that mean ≠ median, and for the vast majority of schools and subjects, marks can exhibit a positive or negative skew rather than perfectly symmetrical data. I found this myself while working on the raw atar calc, when I initially included the average aligned/scaled mark conversions in my modelling it was significantly different to the percentiles, and once I removed them it made quite a significant difference to accuracy - and that’s on the scale of an entire subject cohort, singular schools are significantly more prone to outliers. Band 6 counts are far from a perfect method either (and in an ideal world everyone would provide multiple years of data as that would improve accuracy by a fair bit), but the way they work off a specific percentile is a significant advantage.

A better data point I’d like to see more people include is band 5 statistics, which I’ve found many schools include in their annual reports, as they would aid significantly in ATAR estimates where people have ranks in that range.
Click to expand...
I was thinking about just yesterday. Typically in this case the skew would predominantly negatively skewed.

For example, if we imagine a class which does Subject A, and let the average be 80. Last rank would typically be a very low number, say in the 30's whilst rank 1 will be usually 98~. Even though 98 is a score which in reality is extremely hard to get, the person with a score in the low 30's "impacts" the average a LOT more.

I'd say for most subjects, the median that is the middle rank is 1-2 higher than the average score.

Whilst I agree that band 5 statistics would be very appreciated, I think the average HSC mark is the best "extra" statistic we've got that can be somewhat easily accessible. Great suggestion though!
 
T

TheGrindNeverStops

New Member
Joined
Feb 17, 2022
Messages
8
Gender
Male
HSC
2023
Sorry to revive an old thread, but I'm the person who maintains the site and just saw this thread, hopefully I can answer some of your questions.

Privacy Restrictions
There are 2 main privacy restrictions on the site.
  1. 10 people taking a subject - obviously for privacy reasons this stops marks from being able to be attributed to individual students from small cohorts. This also stops outliers where one student might have done insanely well in a cohort of 1 person and that would make their average really high. Data for schools with less than 10 students taking a subject is not published
  2. The top 10% rule - By law, you cannot rank a school as being in a percentile lower than 90 (ie. not being in the top 10%). In theory, this is to protect schools that are "ranked lower" (there is a very strong correlation between school performance and socioeconomic advantage). The DOE has been very strict on following this rule, they are not even willing to allow, say the top 10 ranks in each subject. Unfortunately, what that means is that some subjects (like the best subject SDD) are left with very limited data. For instance, in SDD, I have the average scores for 49 schools. That means only the top 4 schools can be shown, which is quite limited. However, while further ranks cannot be published, there is nothing to stop somebody from looking up the average score for a school and guessing its rank. For example, if you were to look at Sydney Technical High School (hscscalinggraphs.au/rankings/2022/Sydney_Technical_High_School.html), you can see their SDD average is 86.2, which is not much less than SDD rank 4, who have an average of 87.7. So, you could conclude that Sydney Tech would not be far from the top 4. Average scores are published for most schools, you just can't publish their ranks (but there's nothing stopping you from finding them yourself either by estimation, ranking every school or otherwise).
Median Marks?
I too would love median marks to be published publicly and agree that they are a more accurate measure of performance. Unfortunately NESA does not bless us with these numbers

My school isn't there!
Firstly, I only have the data for public (including selective) schools (as mentioned earlier in the thread). Private schools all publish this data in different ways (if they even give out averages!) and it's too hard to try and compile them. If you go to a public school and can't find your school, I probably have your data. It's just that my in-site search engine is a bit dodgy and doesn't know about your school. Shoot me a PM with a school name if you want me to check for it (some rural schools don't give out any averages, I think this is due to low student numbers).

Best of luck with the HSC.
 
You must log in or register to reply here.

Users Who Are Viewing This Thread (Users: 0, Guests: 1)

Top