Gifted Issues Discussion homepage Forums Identification, Testing & Assessment Evidence for unreliable scores from young kids

I am just wondering if those of you in the know can help me understand what evidence there is that IQ scores from testing conducted when children are young (say from 4) are likely to be an overestimate and will often stabilise to a lower score when a child is older. I ask because I often see that comment here but it goes against the information I have been given from other (also knowledgeable) sources. I hope I am not sounding disrespectful to those of you who know significantly more about testing than I do - I guess I am just trying to straighten it out in my head (I should note my daughter was tested early, but the level she works at supports her test results). I know statistically regression toward the mean would indicate that all things being equal outlier scores should go down on subsequent tests, but it would be difficult for �all things to be equal� with early testing of gifted kids (which I guess would also contribute to how valid the norming sample is - apologies if I am using incorrect terminology). I have been scouring my uni�s journal library for information and can only find information saying testing prior to 4yo being unstable. I have heard Miraca Gross talking about this matter and she said that in her experience results before 6 are often unstable, but are usually an underestimate rather than an overestimate and this was also the experience of our tester, who preferred not to test before 4 but felt that the result from then were likely to be stable or an underestimate (she only tests gifted kids).

The reasoning that both Miraca Gross and our tester gave is that often younger children don't engage as well with testing when compared to older kids and they are often less likely to complete the subtests to the level they are able - simply due to test fatigue, lack of concentration etc. These kinds of things would seem to cancel out the regression to the mean argument because it means you don�t have an accurate position to begin with (I have only a very basic knowledge of statistical analysis so I am piecing together bits and pieces and appreciate my logic may be very flawed!)

Like I said, I tried to track down some studies, however the ones I found were either conducted with autistic kids or �at risk� kids. While each of the studies that I found on autistic kids stated testing in preschool and then follow up testing produced statistically stable results, I didn�t know whether or not that was likely to be the same for gifted children. The results for at risk kids showed a decrease that correlated to the number of risk factors they were exposed to, but this wasn't really the information I was looking for. I found a couple of studies on the stability of intelligence scores in general, but was unable to access them due to a system problem.

I am just wondering if those of you who know such things could shed some light on this for me. Thanks!

Webb, Gore and Amend in A Parents Guide to Gifted Children state that the accuracy of the (iq) test increases from age 3 to 14 - 15 and is quite accurate for children by 10 or 11.

Thanks for asking this! I was suprised when I saw it mentioned in the other thread as like you it's the opposite of what I had previously heard and I am curious.

It's possible that the scores of young kids are more subject to regression toward the mean because there are more random variables affecting their performance (did they nap, did they eat a snack, are they fussy, sick, afraid of the tester, etc. all these affect young children's performance more than older kids.)

Here is a quote explaining regression toward the mean


Source: http://en.wikipedia.org/wiki/Regression_toward_the_mean

Anecdotally, my son has been tested twice. The first time at age 4.5 on the WPPSI and the second time at age 6 on the WISC. His scores (both above 99.9 %ile) were within 2 points of each other. I think kids whose behavior is relatively stable at a young age can be quite consistent.

Thanks Cathy. Sorry to be persistent - I get all that, but to me all those kinds of variables are more likely to result in a reduced score on an early test than inflate it. My understanding (which I am willing to admit is minimal) is that the tests are designed in such a way that luck can only play a small role in the result. A well rested, well fed kid who is fit a healthy and confident with the tester is obviously likely to score better than a kid who is sick/tired/hungry etc. To me it would seem that the kid in the optimal circumstance is simply likely to achieve the score that represents what they can really do. They might regress a bit on a subsequent test in the same conditions, but given the nature of the tests, there is only so much you can fluke. If you're not in that optimal space, you're much more likely to get an underestimate. Which was the point of Miraca Gross and our tester.

I can kind of see how a child might start out strong and then have their development slow down, but I don't know enough about children's development to know whether that is possible or not.

And please don't hesitate to tell me I've got that all wrong - you'll have to forgive me as I'm not terribly sequential and if I'm missing a piece of the puzzle often I miss the whole picture!

What you have to remember when using the argument that young kids may be uncooperative/distractible/tired etc. is that all these conditions applied to the norming sample, too. If you take a sample now of kids testing, their scores will be "inflated" if they are on average less uncooperative etc. than the norming sample, "deflated" if they are more so. There's no reason that I can see to expect children being tested now to be *more* subject to those factors that might reduce their scores than the norming sample were, and therefore no reason to expect their scores to be underestimates as a group.

If we dispose of that set of arguments on those grounds, then what remains is the regression to the mean argument.

What an individual tester sees is going, of course, to be heavily dependent on which children come to that tester to be tested. That could be affected by a lot of different things. For example, a tester who draws testees from a population where high IQs are the norm and only a really exceptional child is likely to be brought for testing (or one, like Miraca Gross, famous for being interested in PG children) is more likely to experience that scores tend to be underestimates than a tester who tests from a more average population.

ColinsMum, I understand what you are saying - the "small child" type variables apply just as much to the sample group as to the individual child.

What I don't really understand is how the example of regression to the mean that Cathy posted applies to something like IQ testing. A multiple choice test sure, particularly one that where everyone answers randomly (which isn't really a test at all!). But an exam that requires knowing the material and being literate, or an IQ test, while I can see there being variability from day to day based on how good you are feeling, raport with the tester, etc, it still seems to me that these are tests based on skill/ability and that you aren't going to randomly swing from the 99.9th to the 50th? Maybe 5% but not 50%!

You can do well on a multiple choice by "getting lucky", but I don't see how you can do well on tests like Vocab, Block Design, Word Reasoning, etc by luck. "Having a bad day" aside I am just not understanding the idea that children tested young will swing back to the mean, or why this is only thought to apply to young children.

Under performing because of being young makes perfect sense, over performing really doesn't. So surely young children are more likely to trend up than other test subjects?

Trending upward on an IQ test might make even more sense for a young HG+ kid, because of psychological factors such as perfectionism that might be overcome as the child gets older, but also because reaching the child's limit on certain tests might take significantly longer than for an ordinary child (and in fact in some cases it seems that kids peter out before showing what they can do). An HG+ young kid might be significantly more able, but without significantly more stamina, and with significantly more performance anxiety.

Yes - I think I agree, which was my earlier point about the norming sample possibly not being terribly valid because these would be issues all younger kids have that would mean the scores would potentially not be very stable in general.

If we dispose of that set of arguments on those grounds, then what remains is the regression to the mean argument.


But this seems to go against what others have said, which is that very high scores in young kids are likely to be very unstable? (Sorry, I can imagine better informed people reading my questions and thinking "what's not to get??") I also had the same thoughts as Mumofthree earlier today - I can see how regression to the mean would work on something like multiple choice, but from what I understand don't you pretty much have to earn your answer in an IQ test?

Hi Lucounu, yes - that's what seems logical to me...