Test Validity

B-APT Validity Studies

The most practical question to ask about any test concerns its validity. Studies have established both content and criterion-related validity for the B-APT.

Content Validity

Content validity is obtained when every question on a test is highly related to the skills required by a job. The B-APT has high content validity for the programmer job because it:

· requires learning a language to communicate with a computer.

· requires that the candidate, using the language rules, write coded instructions

· in a logical sequence in order to solve a problem

· requires analysis and understanding of simple programming specification.

A number of companies justify their use of the B-APT on the basis that it is content valid.

Criterion-Related Validity

Criterion-related validity studies correlate test scores with measures of performance. For example, trainees selected by the B-APT are evaluated by their instructors at the end of training, or by their managers, after they have been on the job for several months. Then the B-APT scores are mathematically related to the performance ratings given by the instructors or managers. The size of the correlation tells you how well the test score predicts the trainee's future performance.

The average criterion-related validity coefficient for general aptitude tests is approximately .30. The following studies, however, show that B-APT validities range to .71. All of them are significantly higher than the average for other tests.

The next section summarises validity studies using training criteria and on-the-job performance criteria. B-APT is not merely a "trainability" test; it also predicts technical competence on the job

B-APT Criterion-Related Validity Studies

I. Studies Performed Using Training Criteria

Company Name	Sample Size	Correlation of Test Scores to end of Training Grades
AT & T	337	.61 TO .71 **

U.S NAVY	279	.70**

CNA INSURANCE	65	.70**

Los Angeles Trade & Technical College	47	47

Blue Cross/Blue Shield of Massachusetts	25	.46
(These companies pre\selected examinees; the corrected, restricted range correlation =0.85**)

II. Studies Performed in America Using the On-The-Job Criteria


Company Name	Sample Size	Correlations: Test Scores vs on-the-job Programming Proficiency	Correlations Corrected for Restriction of Range

American Mutual	28	.55**	.76**

Blue Cross/Blue Shield of Iowa	37	.41*	.49**

Blue Cross/Blue Shield of Maryland	47	.50**	.62**

Federal Reserve Bank of Atlanta	24	.62**	.71**

Federal Reserve Bank of Cleveland	37	.50**	.67**

Florida Power & Light	62	.50**	.66**

Liberty Mutual Insurance	22	.51*	.74**

Northeast Utilities Pennsylvania	22	.57**	.70**

Mutual Life	48	.57**	.70**

Rand McNally	33	.46**	.61**

Republic Steel	27	.56**	.69**

Revenue Canada Taxation	42	.69**	.69**
Security Pacific National Bank	34	.48	.66**

* Correlation is significant at the 0.05 level, that is, there are only 5 chances in 100 that the obtained correlation
occurred by a sampling accident.

** Correlation is significant at the 0.01 level, that is, there is only 1 chance or less in 100 that the result is accidental.

Other B-APT Statistics

In addition to validity information, it is customary to report the data below.


Sample (Aggregate of those in the cited studies)	1459

Kuder-Richardson-20 Reliability	0.96

Mean Score in the U.S	12.60

Standard Deviation	10.01

Standard Error of Measurement	2.00

The Kuder-Richardson Reliability is a measure of the internal consistency of a test. A reliability of 0.96 is unusually high, and means that the B-APT questions are consistently measuring the same programming aptitudes.