Assessment questions are simpler than they look. Here is the whole area in one page.
Reliability, validity, and the standard score conversions. What the exam actually asks, without the textbook chapter.
By Dr. Andrew Wichterman · September 14, 2026 · 5 min read · Updated October 4, 2026
Assessment and Testing has a reputation as the "math" area of the NCE, and students postpone it. In practice, the items are mostly vocabulary and a handful of conversions you can memorize in an afternoon. Here is the area reduced to what gets asked.
Reliability is consistency
A reliable instrument gives similar results under similar conditions. The exam asks you to name the type from a description.
- Test-retest: same test, same people, two occasions.
- Alternate forms: two versions of the test, same people.
- Internal consistency: do the items agree with each other. Split-half, Cronbach's alpha, KR-20.
- Inter-rater: do two scorers agree.
Reliability is reported as a coefficient from 0 to 1. Higher is better. A test can be reliable without being valid. It cannot be valid without being reliable.
Error and the standard error of measurement
Classical test theory says every observed score is a true score plus error. Reliability is the proportion of score variance that is true-score variance, which is why a coefficient of .90 means ten percent of the variance is error. The standard error of measurement turns that into score units: SEM equals the standard deviation times the square root of one minus the reliability. A test with an SD of 15 and reliability of .91 has an SEM of 15 times the square root of .09, which is 15 times .3, or 4.5 points.
The SEM builds confidence bands. An observed IQ of 100 with an SEM of 4.5 means the true score falls between roughly 95.5 and 104.5 about 68 percent of the time, and between 91 and 109 about 95 percent of the time. The exam asks this two ways: directly, by giving the numbers, and conceptually, by asking what the SEM is for, which is interpreting a single person's score, as opposed to the standard deviation, which describes the group. The standard error of estimate is the neighbor: it describes error in predicting one variable from another in regression, and it is the distractor on SEM items.
Scales of measurement and what they permit
Nominal scales name categories: diagnosis, gender, group assignment. You can count them and compute a mode, and the statistic is chi-square. Ordinal scales rank: class standing, Likert items strictly speaking, percentile ranks. The intervals between ranks are not equal, so the median is the appropriate center and you cannot meaningfully average them, which is why percentiles are not averaged. Interval scales have equal units but no true zero: temperature in Fahrenheit, IQ and T-scores. You can compute means and standard deviations and use t-tests and ANOVA. Ratio scales add a true zero: height, reaction time, number of symptoms, and permit statements like "twice as much."
The mnemonic NOIR covers the order, and the exam tests it by asking which statistic is appropriate for a given variable, or which scale a described variable is. The trap is percentile ranks: they look like numbers you can average, and they are ordinal. A student at the 50th percentile and one at the 90th are not separated by the same amount of ability as students at the 10th and the 50th, because the normal curve is dense in the middle and thin at the tails.
Validity is whether it measures what it claims
- Content: do the items cover the domain.
- Criterion: does it agree with an external measure. Concurrent if measured now, predictive if measured later.
- Construct: does it behave like the theory says it should. Convergent and discriminant evidence.
- Face: does it look like it measures the thing. Not technical evidence, and the exam knows it.
The standard score conversions
Memorize these. They come up every time.
- z-score: mean 0, SD 1.
- T-score: mean 50, SD 10.
- Deviation IQ: mean 100, SD 15.
- Stanine: 1 to 9, mean 5, SD about 2.
- Percentile: the percent of the norm group scoring at or below.
Under a normal curve, about 68 percent fall within one SD of the mean, about 95 percent within two, and about 99.7 percent within three. So a T-score of 60 is one SD above the mean, roughly the 84th percentile. An IQ of 70 is two SD below, roughly the 2nd percentile.
Worked conversions
A client's z-score on a scale is 1.5. T equals 50 plus 10 times 1.5, so 65. Deviation IQ equals 100 plus 15 times 1.5, so 122 or 123. Percentile: one SD is the 84th, two is the 98th, so 1.5 is in between, around the 93rd. A client has a Wechsler IQ of 85. That is one SD below the mean, z of minus 1, T of 40, about the 16th percentile. A T-score of 70 on an MMPI clinical scale is two SD above the mean, the 98th percentile, which is why 65 and above is the clinically significant range on that instrument.
A stanine of 5 is the middle fifth of the distribution, roughly the 40th to 60th percentiles; a stanine of 9 is the top four percent. Stanines are the coarse scale the exam uses to test whether you understand that a standard score is a position, not a count. If an item gives a raw score and asks for a percentile, it is missing a step: you need the mean and SD to get z, then z to get the percentile. An item that skips the mean and SD is asking you to notice that the conversion cannot be done, and the keyed answer says so.
Norm-referenced versus criterion-referenced
Norm-referenced compares you to other people. Criterion-referenced compares you to a standard. The NCE itself is criterion-referenced: you pass by clearing a cut score, not by beating other candidates.
The instruments that get named
Know the category and purpose of the commonly cited instruments: the MMPI, the Strong Interest Inventory, the Myers-Briggs, the Beck inventories, the Wechsler scales, the Stanford-Binet. You will not be asked to interpret a profile. You will be asked what kind of instrument it is and what it is for.
That is the area. One afternoon to learn it, one hour a week to keep it.