<p>I know the overall percentile isn't just an average of the percentiles for each of the three sections, but does anyone have any sense of how they relate? Is averaging the three at least roughly in the right ballpark? My daughter struggles on the quantitative section, and I'm wondering what her overall percentile would be like if she was in the 30s for math but around the 90th percentile for reading and verbal. Averaging those numbers would suggest she'd be around the 70th percentile, but I'm just worried that the one low score would drag it down even lower than that. Given that schools only report their average SSATs as the overall percentile, I'm having a hard time trying to get a read on whether we're looking at schools that are realistic for her or not.</p>
<p>Okay, I know I am putting my nose in it with this one. If my daughter were struggling in Math with 30s, I would only consider schools that have a program to work with that issue. The Roxbury at Cheshire Academy(CT) is almost a one-on-one tutor situation and could be a good fit. I would seriously look at the schools’ math programs and NOT have her apply to ones where she clearly cannot handle the workload. This is a tender time for girls, and self-esteem needs to be protected and built up. (IMHO, I would steer clear of schools with 85> as average.) </p>
<p>No, the average of the three is not the overall for her gender. My D’s average of her three scores is three points below her official overall female grade 8 score.<br>
Good job on reading and verbal and good luck. </p>
<p>Correct me if I’m wrong, but I thought the average percentiles were based solely on how the student performed among their peers </p>
<p>Basic comments: </p>
<ul>
<li>Overall percentile is definitely not the average of the individual percentiles.</li>
<li>This is because There are more people who will get one high subscore and modest other subscores than there are people who will do equally well on all substests. Hence average of subscore percentiles at high end is lower than overall percentile will be.<br></li>
<li>Magnitude of difference from average of percentiles is not extreme. You can always average then and add a few percentage points and be close. </li>
</ul>
<p>Nerd threat level red comments:
There is very good data on total scaled scores to percentiles over many years. The data is good enough to allow calculation of a very specific formula for translating your child’s total scaled score into a percentile for their grade and gender. I did this for 8th grade boys, because I happen to have two of them, and the math is as follows below. One caveat is that there has been some recent adjustment to test population scaling, notably in international vs. domestic pools, but I think the math below will get you very close.</p>
<p>Calculate the percentile using quadtratic equation in form ax^2+bx+c where</p>
<p>x = total scaled score (e.g. 2200, 2050, 2345 or whatever)
a = -0.000475732
b = 2.23366315
c = -2522.72174</p>
<p>Results in R2 of .9989 - aka almost perfectly predictive.</p>
<p>Anyone who wants either the math for your specific child’s grade and gender, or wants some more detailed guidance on the math mechanics, just PM me.</p>
<p>impressive beard</p>
<p>Thanks Charger. Excel definitely makes being a nerd easier, and I am glad if it is of use to anyone who like me finds themselves curious about some of the quantitative aspects of the entire process. </p>
<p>@blackbeard: Would you mind explaining how you derived the equation above? I thought that the relationship between individual subscores and percentile must be linear.</p>
<p>@sgopal2, the tests are normed not to an absolute percent correct, but to a running average of the previous three-years’ cohort. It would make sense to increase on’s R^2 by going to a polynomial fit. Note the quadratic component in this fit seems relatively small. Often these tests will not at all be linear, but compressed at the top end. Also notice that the predictive value of the fit is not accurate over the full range; the R^2 must be calculated from low value (1900?) up to 2350 or so. Is that correct @blackbeard? Over what interval is the fit good to .9989? Certainly not below 1900 or above 2350?</p>
<p>I thought Charger was complimenting your facial hair.</p>
<p>@ItsJustSchool - Yes you are correct about the coefficients being calculate on less than the full range of scores, which I should have mentioned earlier. This is based upon the scaled score to percentile data from 70th percentile to 99th percentile. Happy to work numbers for the complete range if anyone wants them.</p>
<p>@sgopal2 - Not going to be a linear function. The reason is that we are effectively trying to write an equation for the cumulative distribution function of a normal population. Before you blackout because that sounds so annoyingly mathy, try this hopefully simpler explanation. Picture a normal bell curve of scaled scores. This is really a histogram showing the number of students who received each score. Its fat in the middle, and thin at the ends. If you now think about showing the cumulative version of the same, that is how many total students got a certain score or below, you will have a chart that goes only up (not up and down) but will go up the fastest in the middle, and slowest on the ends. This picture shows a normal bell curve with the cumulative curve on the same chart <a href=“https://staff.blog.ui.ac.id/sigit.sw/files//2010/08/02.JPG”>https://staff.blog.ui.ac.id/sigit.sw/files//2010/08/02.JPG</a> </p>
<p>You can also see this in the data itself fairly easily. Data set I was using can be found here: <a href=“http://www.admission.org/data/files/pages/2011-12InterpretiveGuide.pdf”>http://www.admission.org/data/files/pages/2011-12InterpretiveGuide.pdf</a>. For the 8th grade girls for example, there are 26 total scores associated with the 99th percentile (flat curve) but only one associated with the 50th percentile (steepest part). </p>
<p>For those of you who PM’d me for specific numbers, I should have them to you this afternoon…</p>
<p>Information for 8th grade girl population is as follows:</p>
<p>a = - 0.000417229<br>
b = 1.946521952
c = - 2171.072073</p>
<p>Unlike boy numbers, I built this formula by using the data for scores = 1986 or 50th percentile and higher. R2 is still .9993, so not meaningfully less accurate…</p>
<p>Thanks for all the info, @blackbeard!</p>
<p>Makes perfect sense. Definitely not linear. Thanks for the clear explanation.</p>
<p>Just posting more grade/gender info per individual request on 9th grade girls. Figure it might as well be captured here in case it is useful to others. </p>
<p>Numbers for current 9th grade girls are as follows:</p>
<p>a= -0.000344648
b= 1.665414245
c = -1910.498846</p>
<p>where equation is of the form ax^2+bx+c.</p>
<p>@Blackbeard, out of curiosity, did you take the next step and determine the percentile value (by section) of a correct answer to a problem? I think that would be a useless data point that I would love to know. If nothing else, for subsequent tests, all else being equal, it could guide on which section to focus one’s studies.</p>