| | | Sometimes, giving the full WAIS III is not possible or appropriate. You may have to give part of the test and estimate the IQ from the subtests given: - there are time or physical constraints you have to work within, for example, a client might be bedridden and not be able to manipulate test items for many of the performance subtests, or, you and the client may have only two hours for a brief evaluation before they have to leave (they can not return for more testing)
- the Full Scale IQ would not be valid anyway, for example, when the client is using substances. I recall one case of a woman I tested who was using substances. She returned for re-testing 15 months later, after having completed substance abuse treatment and having lived sober for 12 months. I retested her and found her IQ jumped 10 points
- only an estimate is needed, for example, when qualification for adult special education/needs services is the issue. The "final number" may be all the referral source cares about. I've done this in some evaluations and used the saved time to administer a memory or career interests test to add a little more useful information to the evaluation
- IQ is not relevant to the referral question, and a screening is all that is needed to rule out Low Average functioning as a problem, for example, in a referral for testing to guide the therapy process
The Sattler Supplement gives you the rationale behind short forms, several examples, and tables to turn sums of scaled scores into estimates of a Full Scale IQ. See pages 1239-1242, and 1275-1284. A little history is in order here. Researchers have tried two main ways to to shorten long IQ tests. - Some have tried to give only the best subtests, and shortening the test by giving fewer subtests. The problem with this is that depending on what you choose you could be picking a subtest that, while highly reliable, also taps a client's weakness. The other problem is how do you decide which subtests are the best? Are they the most reliable for all children? The most reliable for a child's specific age range? The most engaging? Balanced across the indexes or only from one? If you can decide how to pick the subtests, though, this method is pretty reliable.
This method deserves attention, however, as you might have to do this for other reasons. Suppose you have a visually impaired client. He or she obviously can't do the performance tests, and so you are likely to give only the verbal ones and try estimating an IQ from there. - Others have tried to shorten the test by dropping overly easy and difficult items across the test. In other words, you give all subtests, but shorten the test by shortening each subtest. For each subtest, you give a few the child should pass at that age level, and a few they should not, and adjust the raw scores to assume the child would have gotten all the easier ones right. The problem is that by dropping items, you make the subtests shorter and thus less reliable. You also run into floor and ceiling effects, meaning you give some kids credit for easy items they would not have passed, and don't give some kids tough items they could have passed. You end up inflating low IQ estimates, and deflating high IQ estimates, making everybody look average. This method is thus less reliable.
Once you have given the shortened test, you have to figure out how to come up with a score. - One way is to prorate, or figure the average of the subtests/items you gave, and assume the client would have scored the same on all the other subtests/items that you dropped from the test. This way, you can still use the tables in the manual, which is a good thing. However, you may also give people credit for subtests that might have tapped weaknesses for them, and thus inflate their IQ scores, which is a bad thing.
- Another way is to compute a regression equation. These are computed when researchers have a large sample of people's actual IQ scores from a full administration, and draw out from that sample the special subtests/items they want to use for a short form. They then try to figure out the best mathematical weighting of the scores on these specially selected subtests/items to predict the client's real Full Scale IQ. The problem with this is that a regression equation is highly dependent on the sample it is built on, and may not transfer very well to another population or setting.
Tellegen & Briggs came up with a third option that the field has decided give the best reliabilities (.93 for the best four and five subtest short forms) and the most valid estimates. settled on as the best of both worlds. They compute Deviation Quotients. | Computing a Deviation Quotient
| DQ = |  | | where | Sc = | 
(standard deviation of composite score) | | | Xc = | composite score (sum of scaled scores in the short form) | | Mc = | normative mean, which is equal to 10n | | Ss = | subtest standard deviation, which is equal to 3 | | n = | number of component subtests | | and | | |  = | sum of the correlations between component subtests |
(Xc – Mc) is the sum of the subscale scores you got, minus the number of subtests times ten (the mean score for subtests). This means you figure out how much above or below the expected score they are. You multiple this by 15/Sc, 15 being the standard deviation for standard scores, and SC being equal to that monstrosity above. is the sum of the intercorrelations between the subtests. So if you give PC, V, A, S, and MR, you sum the correlation between PC with V, PC with A, PC with S, PC with MR. The you add the correlation between V with A, V with S, and V with MR. The you add the correlation between A with S, and A with MR. The you add the correlation between S and MR. You take the resulting number, multiply by two, add the number of subtests give, and then take the square root of that.
Mathematically it's the same (trust me), but it's a little easier to do this:
| DQ = | (composite score x a) + b | | where | composite score = the sum of the scaled scores you obtained
a = 15/Sc
b = 100 – n(150)/Sc
| | I hope you appreciate Sattler for doing all this math for you and putting the tables in his supplement! You can see in Table O6 that the 10 best 2, 3, 4, and 5 subtest forms have validity coefficients of .90 or better, and the 4 and 5 subtest combinations have validity coefficients of .94 or better. Sattler cautions that if you think a client will have difficulties on a subtest, don't use a short form that uses that one, since you could get an underestimate of their ability. Sattler doesn't go into this, but when you report the test results you obtained from a short form of the test, you should indicate what you did. For example, you didn't get a Full Scale IQ, but an estimated or prorated Full Scale IQ. When you use an abbreviated form, you should not try to compute Verbal and Performance IQs, since these factors are much, much less reliable when shortened. Sattler indicates you can do this if you give a seven subtest short form though. Keep in mind that not everyone approves of doing this, and there are clear cases where you should not. Based upon what you know about the test and it's use, can you think of three situations in which the shortened form of the test would not be good to give? In case you are curious… on these abbreviated WAIS III, I give the following subtests: PC, V, S, A, and MR and look in Table O10 of the Sattler supplement and in column C10 and convert the sum of the scaled scores for these subtests to the estimated IQ score. Here's how I introduce the test: | | "An abbreviated form of the Wechsler Adult Intelligence Scale III was administered here. The estimated Full Scale IQ score derived based on these subtests correlates .94 with the IQ score based on the entire test, where a correlation of 1.00 is considered 'perfect' and very uncommon. This shorter administration is thus very reliable, as well as less tiring to the client to complete, and the subtests chosen here are less likely to penalize the client for poor educational experiences." | | Sattler says it correlates .94 with the Full Scale IQ, but just because I am curious… I have kept track over the last few years of when I gave the whole IQ test and then computed what the estimated IQ would have been. | Case | Actual IQ | Est IQ | Difference | Direction | | 1 | 97 | 104 | 7 | + | | 2 | 79 | 80 | 1 | + | | 3 | 77 | 79 | 2 | + | | 4 | 66 | 66 | 0 | = | | 5 | 52 | 52 | 0 | = | | 6 | 61 | 62 | 1 | + | | 7 | 68 | 69 | 1 | + | | 8 | 74 | 76 | 2 | + | | 9 | 97 | 96 | 1 | - | | 10 | 63 | 66 | 3 | + | | 11 | 91 | 97 | 6 | + | | 12 | 77 | 77 | 0 | = | | 13 | 62 | 66 | 4 | + | | 14 | 55 | 54 | 1 | - | | 15 | 84 | 82 | 2 | - | | 16 | 91 | 95 | 4 | + | | 17 | 74 | 69 | 5 | - | | 18 | 100 | 99 | 1 | - | | 19 | 64 | 64 | 0 | = | | 20 | 63 | 61 | 1 | - | | 21 | 58 | 52 | 6 | - | | 22 | 97 | 95 | 2 | - | | 23 | 61 | 52 | 9 | - | | 24 | 76 | 67 | 9 | - | | 25 | 59 | 62 | 3 | + | | 26 | 50 | 46 | 4 | - | | 27 | 96 | 97 | 1 | + | | 28 | 73 | 74 | 1 | + | | 29 | 77 | 79 | 2 | + | | 30 | 81 | 77 | 4 | - | | 31 | 65 | 66 | 1 | + | | 32 | 68 | 65 | 3 | - | | 33 | 75 | 75 | 0 | = |
So…. Avg Diff would be 87 / 33 = 2.6 points 15 overestimates 13 underestimates 5 exacts As for error ranges, somewhere Sattler gives + 6 points for a 90% confidence interval. You can see the estimate has been dead on sometimes, and as much as 9 points off at other times. | |
