DOCUMENT RESUME

ID 209 248, TM 810 639

AUTHOR Mullins, Cecil J.: AAA OthersTITLE Weighting of Aptitude Components Based on Differences

in Technical School Difficulty.INSTITUTION 'Air Force Rumen Resources Lab., Brooks Me Texas.

4tREPORT NO APARL-TR-81-19PUB DATE Jul 81

'NOTE 20p.

IDES PRICE APO1 /BC01 Plus Postage.DESCRIPTORS , Ability; *Ability Grouping; Academic Standards;

Admission Criteria; Difficulty Level; IliGrading;Military Traiiing;'*Technical Education; *Validity;*Vocational Aptitude: *weighted Scores*Airy ForceIDENTIFIERS

ABSTRACTNithin aptitude areas (Mechaical, Administrative,

Gen 1, a Electronic) in the Air Force, some technical schoolsMuir her levels 'of aptitude for admission than do others (forexample, there are G10 schools, G60 schools, and 080 schools based onGeneral test scores.in th 10th, 60th, End 80th.percentilesrespectively). The schools, however; glare grades on a scale from 70to 100, regardless of the difficulty of the school curriculum. Thissemis that a score of 82 inn. Gan school is" recorded the same as an82 is a G80 school although thi scores in a G80 school must indicatea higher,level.of performance than the same score does in the G10school. This study is an.evaluation of a method of adjustingtechnical school grades issued by schools of varying difficulty sothal: a view criterion is'formed with all school wades adjusted to thesass metric. Thib new criterion, was then used do recompute aptitudeindexes, whicivwet. compared with aptitude, indexes confuted in theconventional meaner. The new aptitude indexes preditted school gradesin a cross4val0.dation sample patter than did conventional aptitudeindexes. (lather/BB)

if

*****************************************e****************************** Reproductions supplied by.EDIS are the best that can be made

from the original docuient. 111

************* *******************************************************t

AFHRL-TR-81-19

A P FORCE

HU

staimmtaiougreFilillgAill/SNA1101111LIONIMUTICIMICATION

ELKOCATIONAL, MISOUNCIS 111PORMA11001

MOTO WRIC)0 TI 1100111OnI he loon asprodumi as

toothed bon the psalm ae atottbs**odginatiq It.kikes d101.16 IMO boon nu* * Improveospeoduttion

toms of *ow apitim mod In thit dem-ount do not MIMI* sntaaklINIEcial

Pes!ti°n "SWWEIGHTING OF APTITUDE COM?ONENTS BASED ONDIFFERENCES IN TECHNICAL SCHOOL DIFFICULTY

Cecil J. IlinsJames A. Eyries

Malcolm Ree

MANPOWER AND PERSONNEL DIVISIONBrooks Air Force Base, Texas 78235

Jul, 1981

Interim Report for Period I June 1980 I April 1981

%pro% ed (or public reteam...

"PERMISSION TO REPROOWE THieMATERIAL HAS BEEN GRANTED BY

°

Attgiffr

TO THE EDUCATIONAL RESOURCESINFORMATION CENTER (ERICV

LABORATORY

AIR FORCE SYSTEMS COMMANDBROOKS AIR FORCE BASE,TEXAS 78235

2

ye'

NOTICE

When U.S. Government drawings, specifications, or other data are used for any purpoe otherthan adefinitely related Government procurement operation, the Government therefr incursno responsibility nor any obligation whatioever. and the fact that di, Government haveformulated, furnished; or in any way supplied the said drawings, 'verifications, or other datais not to be regarded by implieatica or otherwise, as in any manner licensing the holder or anyother person or corporation, or conveying any rights or permission to manufacture, use, or sellany patented invention that may in any way be related thereto.

This interim report 'his submitted by the Manpower and Personnel Division, under Project7719, with HQ Air Force Human Resources Laboratory (Ana Brooks Air Force Base, Texas7823.5. Dr. Cecil J. Mullins (MOAM) was the Principal Investigator for the Laboratory.

This report has been reviewed by thi. Office of Public Affairs (AA) and is releasable to theNational Technical Information Servise (NTIS). At NI'S. it will be available to the generalpublic, including foreign nations.

This technical report has been reviewed and is approved for publicanoit

NANCY GUINN, Technical DirectorMaispower and Personnel Division

RONALD W. TERRY, Colonel, 'USAFCommander

-

7Unclassified 4

SECURITY CLASSIFICATION OF THIS PAGE (When Pate.EntetA)

REPORT DOCUMENTATION PAGE READ INSTRUCTIONSBEFORE COMPLETING FORM

'1. REPORT NUMBER

AFHRL-T11-81.19 ,2 GOVT CESSION NO. 3 RECIPIENT'S CATALOG NUMBER0

4. TITLE (and Subtitle) . S TYPE OF REPORT & PERIOD COVEREDInterim

WEIGHTING.OF APTITUDE COMPONENTS RASO ON I June 1980-1 1,pril 1981DIFFERENCES IN TF,CHNICAL SCHOOL DIFFICULTY

9 PERFORMING 01G. REPORT NUMBER

. AUTHOR(a) a CONTRACT OR GRANT NUMBER(a)Cecil J. MullinsJames A. Earle;Malcolm Ree

. .,a, PERFORMING ORGANIZATION NAME AND ADDRESS SG. PROGRAM ELEMENT. PROJECT, TASK

Manpower AREA I WORK UNIT NUMBERSand Personnel Division. .' Air Force Human Resources Laberntory 62703F

Brooks Airforce Base. Texas 78235 . 77191804..

IL CONTROLLING OFFIEE NAME AND ADDRESS 12. REPORT DATE

HQ Air Force Human Resources Laboratory (AFSC) July 1981 . .. ..

P'ooks Air Forte Base, Texas 78235 13. NUMBER OF PAGES'C.) a 26.

MONITORING AGENCY NAME 5 ADDRESSOI different from COMfOliklt °MVO IS SECURITY CLASS. (ot Mt* report)

. I .11elassified It.

, 15a. OECLASSIFICATION/DOWNGRACIING ..SCHEOULEk

TS71:7-4STRIBUTION STATEMENT (of this Report), '

- .. .- Approved for public release: distribution unlimited.

eee.

cr. DISTRIBUTION STATEMENT (of the abstract Attired in Block 20, if different from Report)

' .

II. SUPPLEMENT AIRY NOTES'

0

' -HI. KEY WORDS (Continue on reverse aide if necessary and identify by block number,

aptitude. area., .

personnel testingaptitude index'.selectiontrm;d Services Vocational 1,ptitude Battery (ASVA

classification...

.

validityA-

20, ABSTRACT (Continuo on revere* aids if neceasery ancridenttly by block number) )Within aptitude areas (Mechanical. tdministratiee. General. or Electronic) in the fir Force. some technical

sthools require higher levels of aptitude for admission than do others (for example. there are GI scluiols. 1;b0en.hools. and C80 schools based on t;eneral test scor., in the tOth; (10th. and MI. foecentiles re.pectiely). theschools, however. giee grade. on a Cali! from 70 to 10 0. regardle.e. of the diffid ulty of the chool curriculum. Thismeans that a wore of 82 in a CIO whool i rerorded the came as an 1(7.1:1 a G8(1.ch:eol, although the .core in a 1;80school mum indicate a higher level of iwrformance than-the atn .core doe. io the 1:10.rhool. V. lien %al dult,. arecomputed across an entire aptitude area. the different meanings of idt:ntical number.' acne,. -.chew!. of %Ivy ingdiffieulty must confound the re.tilts, I

I "R" 1473IAN?) EDITION OF 1 NOV 65 IS OBSOLETE

0

I etcla..ifit.d

SECURITY CLASSIFICATION OF THIS PAGE (When Data Saftrad)

4

4

Unclassified

CIO

SECURITY CIASSIIIICATION OP THIS PPOICfrkoi Data Entovd0

Item 20 (Continued):

'this stud) is an evaluation of a method of adjusting technical school grades issued bdifficulty so that a new criterion is formed with all school g' ides adjusted to the .ame metric.

.was, then used to recompute aptiludebindexes. Which were compared with foitude indeuconventional manner. The new aptitude indexes predicted school grades in a rioss-validationdid the conventional aptitude.indexes.

k

5

.1

Unclassified .

sehools of vailingThis Ile% criterioncoMputed in the

sainpl; better than

SECURITY CLASSIFICATION OP TIAI4 P4011(1h44 pate Entoted)

A

I PREFACE

This study was conducted under Task T71918, Selection and Clasaification Technologies. Theresearch focuses on the development of procedures and techniques to refine and improvemeasurement devices used in the Air Force operational testing program.

,7.

This work represents an attempt to refine the,aptitude indexes of the Armed Services

Vocational Aytitude Battery (ASVAB), thereby improving their predictive accuracy andconsequently the utility of selection measures. This effort supports the sulithrust area Assesemept of .Personnel Qualifications, under the major thrust area of Manpower and Force Management.

/

e

..

..

6I

.4

.

4.

J6

)

TABLE OF CONTENTS

. PageI. Background and Introduction 5

. IL Approach. 8....

Sample PopulationPredictor/Criterion Variables 8Method 9

M. Results and Discussion 10

IV. Conclusions and Recommendations 16

References 17

8

UST OF ILLUSTRATIONS

Figure * Page1 Schematic representation of depressing effect of similar criterion

range on overall validity coefficient computed acrossschool requiring different levels of aptitude 6

2 schematic representation of higher validity coefficientattainable if different level schools are placedon same criterion metric by adding constants

LIST OF TABLES

Table PageGroups by Aptitude Area and by Entry Level 8

2 Comparison of Uncorrected Validities, Three Prediction CompositesAgainst Technical School Final Grade 11

3 Comparison of Corrected Validities, Three Prediction CompositesAgainst Technical School Final Grade 13

4 Improvements in Prediction by C2 Composites, Base and Target Schools Compared 15

WEIGHTING OF APTITUDE COMPONENTS BASED ON DIFFERENCESIN TECHNICAL SCHOOL DIFFICULTY

I. BACKGROUND AND INTRODUCTION

C

The use of the official aptitude battery (called by various names over the past three decades) forselection and classifications of Air Force enlisted personnel has always taken the form of computation andinterpretation of four or mode Aptitude Indexes (Al.) (Weeks, Mullins, & Vitals, 1975). The use of A,Isappeared in the first Air Force aptaude battery (AC-1A). It was not administratively feasible in 1948 toproduce a unique composite score for each Air Force job, but it was assumed thit differential aptitudecomposites were "desirable. Job clusters were developed on the basis of subjective judgment and jobanalysis data. Through study of test results, scientists formed clusters of tests (Al.) which were reasonablyhomogeneous ins, many and predictive of success in schools in the separate jol? clusters.

During succeeding year:, various changes in composition of the Als have been made, mostly byadministrative fiat, sci that at* present timg?"-he current enlisted aptitude battery produces four AirForce AlmMechanical 1M), Adminisiratiie (A), General (G), an9Electronic (E). Along theway, a greatdeal of research has been done on the enlisted aptitude_bititery-, but few studies qUestioned theeffectiveneis of the concept of M, A, G, and F. aptitude-indexes or explored novel ways of weightinginbuilt* to Pioduce the M, A, C, and E composites. This study addresses the utility of a different method

weighting the M, A, G, and E composites.

toricidly, subtest weighting has been accomplished partly by science and partly by artistry.Through various multiple cotrelational techniques, an optimum weight has been derived within each AirForce Specialty for each subtest score against final technical school grade for that specialty. Then the setsof weights for specialties have been scrutinized within a aptitude area (say, M), looking for an.inimal set of predictors which consistently exhibit 're non-trivial- weights across the entire area.When such a set has been found (three or fourpedictors), the aseish'I are all rounded to 1.0, and slainmultiple correlation coefficients are com?uted between school irides and these unit-weighted predictorvariables to see if the validities are holding up after ronversion from optimum *eights to unit weights.Ordinarily, little is lest by converting to unit weighting (see Wainer, 1976).

One problem, however, has been recognized with this system. Different schools within ealt aptitude.

area require different AI level- to qualify,for entry. For example. some A schools require only score of40th percentile for admission.While others require the 80th percentile. Both Schools, however, gi e gradeson the same apparent scale. from 10 to 100, even though the A80 school is undoubtedly mu It more.difficult that the A40 school. Therefore, a final school grade of 82 would refer to i Weser arcomplishrnentin the MO school than it would in the A80 school. When validitiesire computed and predictor *eightsassitmed across entire aptitude areas regardless of school level (see Figure-4), sorrl method is needed for

taadjusting school grades in individual schoids upward or dOwnward as a function of the prerequisite velsof ability (MO, A50, . . . Alle). Such a method would ensure that graduates of A40 and A80 schools vecriterion doses based on the same metric. In short, if it could be done, predictor weights for subIestswould be more accurate. and Al scores could be computed which would be more ef.icient than they',,erenow. The problem. then, is how to estimate what the school grade of the A80 students would have been ifthey had taken The A49 course and if there had not been a ceiling score of 100 on school grades. .2 .

%N.

100

4

IllI

40 80

Entry Prerettoisffe

Line of beat fit for each le,eiof achooisLine of best fit for entire aptitude area

A MO entrance prerequisiteB 010 entrance prerequisiteC ADO entrance prerequisite

Figure I. Schematic representation ordepressing effect of similar criterionrange on overall validity coefficient computed across

school eetiniring different levels of aptitude.

--When restated x. this form, the problem almost resolves itself. %he solution is to find a constant that

can be added to the school gradei of AIXI students to reflect the difference in difficulty between the A40and A80 schools. Such a constant should improve the situation in the manner depicted in Figures I and Z. ,

,The es on of this constant requires only that the mean school' grade of the A40 school be knownand that an estimate can he made of the mean school grade the A80 students would have earned if theyhad ceded the A40 school and if the 100 score ceiling were removed. Such an estimate can .be maderemenebty well by computing the hert41 in the A40 school from available predictor information. This Alis then mid to predict the grades of members of the A80 group. The difference between the mean of theobserved criterion grades of the A40 group and the mean of the predicted grades of the A80 groupprswitiorthe required constant This constant is then added to the criterion grade of each subject in theASO poop to pined' a raw criterion metric go that grades of all students (both A40 and A80) are arrangedon the same criterion scale.

The formula to derive the new criterion K score is as follows:

(K.*

IG + - )

where

K. the transformed grade score of person j

G. the observed grade score of person j

6

a9

C

a

.

17B

100

,70

/_ / /

C/

///A

/,

0 r

40 80

Entry Prerequisite

= Line of belt fit for entire rptitude areaA - MO entrance prerequisite

- MO entrance prerequisite- A80 entrance trerequisite

Figure 2. Schematic representation of higher validity coefficiefitattainable if -different level ichools are ple.ed.on same,-criterien metric by adding constants.

the mean of the composite-scoresgenerated for stu dents in the Base group (the groupin whichuhe prediction composite is generatedi.e.. the A40 group in the aboveexample).

the mean of the composite scores generated for subjects in a Targo group by applyingweights developed in the Base group (the Target group is the group to which thecriterion grade correction will be applied. In the above example. A80 would be theTarget group).

Whenibe scores on the K criterion have been computed. the siltation depicted in Figure 2 will havebeen achieved, and an adjusted criterion wilt have become available for use in developing new weights forthe available predictor variables. The new weights can be used to establish a new aptitude compositewhich may reasonably be expected to predict success es throughout the aptitude area. disregArding level.better than any set of weights computed in the conventional way.

Two sets of weights ire computed. The first set cornesfro' m predicting the actual grades on just theA40 group and is done only as an intermediate step to determine the constant used to adjust the grades ofthe A80 group. The second set of weights comes from predicting a combination of the actual grades on theA40 group anfl the adjusted grades on the ABO group. This second set of weights defines the new aptitudecomposite.

0

*

1

7 104

f

0.

)

.,,

After the new.weights (against the K criterion) have been established and composite aptitude scores/

have been computed for all students in the study, it is necessary to check empirically to see whether thenew composites really do predict, actual school grades better than do the old ones. The objective of thisstudy was N3 develop new weights for aptitude composites computed from K-criterion scores for a sampleof the population and to cross-apply these weights to another sample.

- :

11. APPROACH

Sample Pokilation :s

. . , .

vThe _sample consisted of all airmen entering the Air Force between Jaml' : 1977 and Septenibt.

1979. on whom subtest and Al scores on Armed Services Vocational Aptitude Battery (ASVAB) Forms 5.6, or 7 and numerical technical training final school grades were available. Schoolofailtires were omittedfrom the sample, as well as all subjects in schools w-1- 're the total number of graduates during this 2-yearperiod was less than SO. Total N for the sample after all necessary deletions was 88.199. Of these. 19.715were graduates of schools requiring multiple aptitude prereqiiisites (e.g.. E80 and M60). and 68.484 werefrom 119 schools with only a single prerequisite (e.g.. A60). The 19.715 subjects in schools with multipleprerequisites were arbitrarily called X, and all computations and data Thiiiiipulations applied to the M. theA, the G. and The E subjects were also applied to the X subjects (even though this group consisted of M. A.G, and E subjects intermingled).

The subjects in each school were randomly divided equally into a computing (C) subsample and across-validation (V) subsample. Then, within each sttbsdmple. schools were combined to form the groupsshoWn in Table I.

Table 1. Groups by.Aptitude Area and by Entry Levelasoaz_

Group N(C+V) Group i N(C+V)

-,,

M40 8,3958 G40 3,530

I M50 8,079 C45 14,271a

A40 7,259a G50 .154

A50 224 G60 8,710A60 2,251 G65 121

A70 205 G80 773

A80 1,204

Group N(C+V)

E50 1,852E60 1,134E80 10,3228X40 5,078X50 1,266X60 13,3718

'Base group. All other. are Target groups.

Predictor/Criterion Variables

The following vari,Ibles were available (or were computed) on each subject:

I.2.3.

Technical school fibal grades, graduates only.ASVAB subtest score Numerical OkrationsASVAB subtest scoreAttention to Detail

,1

i 8

1'

4. ASVAB subtest acor Word knowledge5. ASVAB subtest = pre Arithmetic Reasoning6. ASVAB subte: scoreSpace Perception7. ASVAB su t scoreMechanical Comprehension8. ASVAB s test scoreShop Information9. ASVAB iubtest scoreAuto Information

10. ASVAB subtest woreElectronics Information11. ASVAB subtest scoreGeneral Information12. ASVAB subtest scoreMath Knowledge13. ASVAB subtest scoreGeneral Science14. Mechanical Al, as conventionally'derived.'15. 'Administrotive AI, as conventionally derived.'16.'- General Al; as conventionally derived.'17. Electronic AI, as conventionally derOed.'I8-59. Educational ,,,ariables. These variablei were dichotomou . scored 1 if the subject had

successfully completed a specified public-school course, zero otherwise.60-64. Prediction composites CIM. CIA, CIS, CIE, and C1X. computed against the K

criterion using only the ASVAB subtest scores (Variables 2-13).65-69. Prediction composites C2M, C2A, C2G. C2E. and 02X-computed against the K

criterion using the subtest scores and the educational variables (Variables 2 -13.18-59).

Method

The K criterion was computed in the C subsample of the M Base Group (M40 schools), and applied inthe Target Groups (only one in this case) of that aptitude area to get the constants for correcting the finalschool grades of each subject so that all members of M schools were placed on the same criterion (the. Kcriterion) metric.

This procedure yielded a single criterion for all members of the M aptitude area, regardless of level.The kvejes..were then combined, and within the M aptitude area, another 113 was computed in the Csubsample; this one between the K criterion and the 12 predictor subtest scores taken as a of Using theweights emerging from this exereiv a new Mechanical AI score (called C110-wer-Odirated for Pallsubjects in all cross-vididation subsamples (A, G, E, and Xlas well as 14)..-116-completed the developmentof the C1M composite. The same procedure was repeateiin t C, E, and X groups to generate CIA.CIS, CIE, and C1X for all subjects.

The procedure described in the p us two paragraphs was repeated, this time using the 12 subtestscores plus the 42 educational v es as the set of predictor variables. The prediction composites using

t all these predictors we ignated as C2M, C2A, C2G, C2E, and C2X.

At t t, three different sets of Ala, or predictor composites, were available for comparison in thea idation sample; namely, the four composites generated in the traditional way (M. A, G, and E),

---- ----the five CI composites generated using the K criterion and the subtest scores only (C1M. CIA, CI G, CIE,--_------ and C1X), and the five C2 composites generated against the K criterion using the subtest scores plus the

'In this study. the M. A.G. and E aptitude composites were recomputed and used in raw score (not percentile) form. Conversionproblems with ASVAB t and 7 would not affect the results of the study.

12

educational variables (C2M. C2A. C2G, C2E. ind C2X). Validiry comparisons ere ,made in the Ysubsample between the standard Ala and the CI and C2 composites to determine wh ther or not the Cland/or c2 composites improved prediction of final school grades in individual schools, and if/so, howMuch improvement occurred.

-

III. RESULTS AND DISCUSSION I

Validity coeffiOento against school grades were computed within The`"of the 119 schools: Thuueorrected lalidities of the Ms, the CI crposites, and the C2 composites are shown in Table 2. The

4mmevatiditie conlected for attenuation by selection (Guilford, 1950, 1 ninlac 113.2kend 13:31,. 3 9)*-Pu

are shown in Table cl. The following observations are obvious from T. :de 3:, I i ,

.

.

1. There is verYlittle different* between the Cl and C2 composkes. Validities averaged (usink R to.Fisher), Z tree lormition) "cross all schools were .59 for the CI composites and .60 for the C2 composites.To improve validities by an average of only .01 is not worth using 42 additional predictor variables. (the,education variables). For thz rest of this report, comparisons will be nude only between the conventionalAls and the Cl composites. 1

...

2. There is worthwhile improvement, overall, in the predictive efficiency of the Cl composites ascompared with Ala computed in the traditional manner. As mentioned in the,previous paragrph: the /overall average validity of the Cl composite, across all i 19 schools, was .59. The average validity of the Ala

4bitacross all schools will .50. It s' .ild be noted, however, that the improvem t in prediction usi-e, the GIcomposites may not be entirely attributable to the new way of computing th Cl composites, using the Kcriterion approach. There were at least tiro other differences between\ e formation of the traditional Al:and the CI compor 'es. First, all ASVAB subtesta wers,used to fo he Cl composites, whereas onlyselected subsets of subtest scores are' used to fq115 the traditional Ala. Secoltd, the subtes* I scores,comprising the its were unit weighted, whereas the CI composite was formed by optimal weighting of

4 '-12 subtest spores. Experience indicates that, in a cross-valiestion sample, optimal weights produce very". hide more than unit weights and that, at !rest in most situations involving a large predictor set,

Only a very ew variables have we ificantly different from zero. From a practical standpoin5. theimportant fact is that comrade. ....imputed in saner of the CI composites are more efficient inpredicting success of airmen, for whatever reason. Still; it i rtani to understand more exactly why theCI composites are superior to the Ala coin'. - .c4t in the usual mann?, reanalysis of these data will bedone to muesli for the variance which could possibly be introduced. by o al weighting and largerpredictor seta in forming the CI composites.

3. The validities of CIM, CIA, C1G. CIE, and CIX are all very similar, regardless of what is beingpredicted. For example, there is very little advantage in using the C1M composite to predict success in themechanical area; CIA, C1G, CIE, and CiX all do about equally well. This is an interesting finding. It %seems to asgur that success in one area is similar to success in other areas. Also, differential prediction byfeats cif 'various "factors," which. most researchers have been pursuing. may be. as McNemar sugge" A

(McNemar, 1964), lamely illusory. Certainly in this study, where no artifijal controls were imposed onthe selection anci weiginine of subtest scores, there is little to choose among the Cl M, A, G. and E ...-

composites, whatever one is predicting. Average y,andities of the various composites are given, by aptitudearea, in Table 3.

differences appear among the M. A. G. and E Ma computed in the traditional manner.although the sele-tor composite is sometimes not the mos efficient one; probably beciniie, in the past,

iO Ififfer-Tcs among the Ala were sometimes forced even though some overall validity was lost. Even theseconventional Al.. formed in a theoretical framework rationally designed to maximize differential validity.are not generally very convincing in substantiating differential prediction as a practical goal of testconstriction.

10

13 ,

VI

I.

ca4

J

14

u1144,Gullule INS

S*AA4tA74A3p*RA44PS3r,UVZAVAg UOSEMAX%4.7.MA

SgTi..7.44FAA4Mr**ASAWAIIR=E4

L34.:4a1,T4=7,A4t4,743t7,g7PPSW7,4V17,AVA

2.74.12:224%43:3233,74.7,22.1.7=11,1;

SAA4,%,W4144"41t4A%A44A*UA4tViiA3M7,

34.1;t24AR4M414-X3S.7,13AAP,%:;$21;;=R

2,4A.%RAX*%4:it4$104A:1441432W4M,7

2=4427,4;A*44 4R4 SAN%7,424AAFAM

UM4105122442.142V =MI4 ZS

4.74M4U4X14.14X4A4474A;$4XWAAgc%.A

MAVt,744;gia,StIZAP% US.1443*AA44

SURAVA.1;4**AAR-MAX:7543n2A

4o44saar4usaram*4ummuz4a1

Umfamo lamm

.4..1.31,DUILM.4.44,1204WM.4.,12

mama .04none4404 14414!9

7,43MR3AL-7,,V4 4aA

.77,SZMSnSg",74.1gZ,

R44VA444.4A447,3

0.e.444---PCMCTMA.d.u142-015u1M..u1M.,PIV..

32g3X=443323.

VASSPAMM%*%4$.S

44M32$P4W;A4L";

4 *ge;44e72.72AM2

.

surnszuzuzs'

AA$U*R8AfAMAX

,..gosegA_ M

TrtftflTMDTp;a2-ANA-E00-

x-IIRISSOMMed atavisilLOPP

*:21241-4'Aa4M22437a7.,*g4

*273WZS444N34.7!il4A7,7

pt449404.0nmn....ei44040m1441. a

4TMAAtg:;$7,!;i1.4.7.4g4

7q2.V.,r04"M447U7-PAM7

32M4424,7*.V.;3A2M43*

Vc17,44n2g444.74A.4AM4A4

4,24"4122Z4tASNA2%2ASS

.727,3gi274447,4424g,124

nSM%44417,A44%z.1%.741

ll4 001Mg44 4C4144441--14t1.1.15074441MMM4.4.44444.4

ie,t 175.

t2M-AA=20M47.AF.0%.7,V4r.

P*744?44245Mgin-A23e8A

-

11111111147117rP60...0400004026...000...0044000.0.000.04OtOW ).00400.0,40e

Table 2 (Continued)

ConventionalComposites

Cl C2Composites Composites

School. ' M A GE M A G E M' A G E X

G Schools (Continued)811X0 (G45)8 23 21 33 27 37 37 37 __---36 37 38 37 38 36 38

811X2 (G45)B 26 20 42 33 45 45 44 46 \ 44 44 44 44 14

902X0 (6601 29 42_ 49 35 48 52 47 , 47 50 49 53 ' 48 48 51

902X2 4G60) , 45 28 43 45 56 53 5 54 55 .55 51 56 , 54 54

903X0 *601 46 37 -46 54 61 57 9 62 60 60 55 58 61 60,

904X0 (G60) 29 22 41 37 461 51 42 47 50 49 si 44 49 54

95X0 (G60)_

38 44 47 49 59 57 62 58 65 59 63 65

406X0 (G601°- 09 28 37 20 33 43 29 35 38 31 42 28 34 36

914X1 (G60) 30 43 48 28 44 49 42 41 46 43 52 42 41 46

915X0 (G60) 17 32" 30 15 35 39 37, 32 37 36 37 37 34 40

922X0 (G46) 28 -41 45 27 49 54 49 46 49 49 52 50 - 47 50

981X0 (G60) 35 35 50 40 51 53 50 52 53 50 52 50 52 50

982X0 (G60) a 26 19 32 27 40 43 40 40 42 40 48- 41 42 44

E Schools275X0 (E60) 49 31 58 54 / 64 66 66 64 66 65 64 65 64 65

302X() (E80)g 15 37 21 27 1 501 41 48 47 47 53 42 53 56 57

303X1 (E80)13 35 34 35 40 51 49 50 52 51 49 44 51 51 48

300X2 (E80)B 55 47 76 64 81 79 79 80 81 78 83 77 82 83

303X3 (MB 24 25 39 29 49 47 48 49 50 48 47 49 54 50

304X0 (DO 30 19 29 36 49 13 47 49 48 . 52 45 50 51 51

304X1 (F-801B 34 34 41 39 56 52 54 56 56 57 54' 56 56 58

304X4 (E80)B 25 31 42 47 57 57 54 61 60 57 55 57 ''62 60

304X5 (E80)B 22 20 4i 34 39 ;40 37 41 40 40 -41 38 43 42

304X6 (E80113 32 51 50 43 62 62 59 &I 64 64 63 61 67 66

305X4 (ENO 23 27 42 31 44 48 44 47 48 47 46 45 48 47

306X0 (E80)B 28 23 36 37 49 45 45 49 50 53 44 46 51 51

306X2 (E80)B 22 38 41 45 57 56 52 60 58 59 58 56 62 61

307X0 (E8010 22 12 27 23 42 38 41 40 42 43 37 42 42 42

)16X0 (E80113 28 ,21 33 33 45 42 43 46 4.5 46 43 46 49 46

317X1 (OMB 28 09 33 32 54 48 51 50 53 54 47 52 52 54

316X2 (D O)8 15- 42 47 48 58 62 53 65 61 ".58 62 54 64 63

3T6X3TE80)B 35 24 27 28 40 37 47 39 39 40 30 45 40 39,

321X0 (E80)0 08 23 39 35 . 44 52 42 49 ,50 43 41 .41 46 45

3t I X I (E80)11 31 26 40 35 54 52 55 52 53 51 51 50 SI 51

321X2- (E80)B 21 19 34 47 44 42 40 48 46 44 42' 46 50 48

322X2 (E80)8 15 09 32 39, 39 38 34 42 41 42 38 35 44 43

324X0 (E8011 31 22 32 45 50 46 46 54 51 53 52 48 58 59

325X0 (E80)11 20 22 29 28 42 40 40 43 44 4.5 45 44 49 48

325X1 (E81011 20 1 _ 27 28 39 37' 36 41 41 42 37 37 43 43

326X0 (E801f1 41 48 " 53 52 71 72 71 73 72 67 61 69 69 66

326X1 (E801B 24 18 34 -34 43 42 39 45 45 42 38 39 44 44

326X2 (E1M)8 37 :6 26 35 47 36 49 44 44 5i 41 53 48 49

328X0 1.011 22 19 , ' 28 32 45 40 44 45 45 16 38 44 48 46

328X1 (E80)B 25 29 39 41 53 51 48 56 55 53 47 49 MY 55

328X3 (E80)B 37 22 39 48 58 54 55 60 60 60 56 56 (3 62

32814 (E1101B 20 '23 40 40 49 49 46 54 52 51 49 47 57 54

341X4 (OMB 31 06 ,-13 32 39 19 33 40 10 46 26 36 49 48

341X6 (ENOS 24 42- 31 39 47 44 44 81 48 52 15 46 55 55

362X1 (E60) 31 18 32 35 48 41 45 48 47 46 38 44 46 45

362X2 (E60) 20 56 57 52 57 65 55 62 62 60 62 57 63 63

362X4 (E60) 42_ 27 :58. 43 68 03 67 66 67 67 64 66 66 66

403X0 (E80113 29 28 50 .51 01 59 57 64 02 60 43 57 62 57

404%0 (E60) 57 53 48 58 65 56 65 65 61 66 57 68 64 63

404X1 (E60) " Ili 14 49 13 44 33 48 31 36 42 34 47 32 35

423X0 (E50) 3-4 19 43 43 52 5t, 50 44 SS so 54 51 52 56 56

463X0 (E60) 44 43 59 49 61 bg 62 (h4 65 62 58 62 63 02

341X0 (E34) . 46 40 46 ' 49 61 '' 56 40 SO 60 62 57 61 60 60

542X0 (E50) 71 22 45 67 75 00' 74 74 71 77 64 75 75 75

542X1 (£.50 40 46 55 48 61 60 61 61 60 64 06 65 64 ('0

%elect. or Al for each school in Bold.mbschnal points hire been omitted front correlations to save space.c"B" designates Base schools. All others are Target schools. .

12

15

;.;

6;71827iSiWi4$MH44Sk244EU

f.74MTEMS:AU-743M.7;MAL3SgA2AA'g.,,,,,.,

F..5.$$M6,M417,=:I.T,M44*2$M.7MPMAI

fi 01%42=%=Ainti

'MMV:4,765A2AD;3A2

g34%2I4-1.47z4A4AA

%2425;4EgEag.%;:at;SVA*4

,4V.".2P.TyM5447,4Mr;r27,274:::1

.7AWZAiWnAm94:,71V4

SziAz*.T,MMW4M$%$15;$%4*%4AMggScA 74;%.5,,T7pRA=la34 SR22422222$2342$1424

ISSMS$M$A1:;ARAR*374A3SUISR4,71 22$222Z22343z 4 AASIAIA.574NMg4s.cts;IM

2S:27337,MSSIta%703MIMUMASAMVX Vhf mr4S5M;11.7 341"' AAP24221=;AIV44.7.MV.

SV:41-M.147,A4MA4CAA4*MV.1AAM*Mztg M 7.=2Are;n2 E;V;..7 2210tS,,C1;*1svp.1442MMMSA4MMSM*4311$M7AAMMM AMR2=-9.14$ 1U&S

..'iS.7,12MMM4*M$4MSM4AISSA44S7ga, M SMSSIAMM aarm 7, Mg22A22.2.7274422.172%.122

21'4i.7,V;,'Mmg:43W-7534:41*'.;,ssMA4WA 5 t2M222E0 2414 21

$..04$.4.00,0040001333*-91..ftdht.PWC$0.0.0.r.P,Owlammoolmw .140.4Mettel4Adtft Z4S.VIUnS4ue4a an en en es1.-0 it3 .=.G -0 an ." an ...T. s."..

M4AAS44Mn444,5.Gt4APSAAM:PUPV S !IUM:734A:43474.1

!AASAALV441114414SMX**USAm4tzg14Mm .7:1,7iStia2VA.7.4z 2SR2422-2423$74=2:;222

U7agIVAw;4MV.AAPA'zv.2X14r.14AUIMM 84;U:34t;$428 2 APAMI;;VPPA;M14A=MF.:4MA

$24444MIZ4M25122.72=SSMUM222 Ax;4,%.7isnam7iv SMAM?%5*A$MAX*1:43.4E4

$0 $00$01

VteoPaROMPAsaleaUbOui

I n ti

f Iiii111111

I

1111111011 ill/IifIBBALUV

tva..tvEwtutasewet4t4etval=mitistortterttagstm=

e ntmnummlertlrivtsymazsrlottm=musiciavellutt

zr.lanwetattrneeA*tstommaccum* NAVVNWV*tt$413

t mutrtmattstnntstmultsamstntststrttzmuntt

znewtt-=amavarttatrestttaecserwrzmeerscs;

L mrJE:2,41:mliztltztasecvmstmscsmataumsturgastsvme.xsG mes..,..vlitavertmeneetemerszersefesitzawseesealle

-izetatutv;sitstatrtsesttattetuntastrosatemut

s,?..virvm4saesenssztamtsamettes;:e4natttvuttvuts

4

atesamftg:I:lemseezIntwvvvliEeswistray.testmEz

:!valgwesteestetVt%W=TateNVeVIWIetStrVStegW8IttS

vms:2$*=&tsmettlisstetratsteletecvvmsstatu..zeeste.T.

t antatcsavot2t,t,gstutstrorttsgs:ramievzttnttt

2.me=amlswe2mtvv.1.04vm...ziestvesnamegsa;c4etz1

.* =*smitF.artew4

e 2tteite.?mievuszt

Vettrtettattt

tEettteVeSVtt

ntrgtStSt%t%

t tSttattSWIttt

P =n avann#EtiliiinOtiWitigietnitaagirdwfitn'

tttlatgreVetttevtntswev.1#%

a

z

L1

1 12

f*

5

a

0.

ean

Whop individual aptitude areas, individual levels, cod individual schools are considered, the Kcriterion technique finds even more utility. The average of Cl' composite validities for M schools is .55; theaverage conventional M aptitude index is A4 (see Table 3). CIA avenges .55 for A schools, whereas theaverage Al-A is Only .2& The average C1G (for G, schools) is .55, while the average AIX is .52. Finally, tl)eaverage OE is .65, compared with an average AI-E of :57. Certainly in the M and A areas; the Cl

'composites are superior to the Al composites. In the ( area, the CI composite is slightly better than the AI.and in the Electronic area; the difference is well worthwhile.

The largest improvement is obviously in the A schools, and a close scrutiny explains why. Of the 13 Aschools, the Al-A composite yields the least prediction of all the conventional AI tomposites in nine ofthem (69%). In fact, in every one of the 41L schools, the conventional Al-G appears to be a better predictorthan Al-A. In no other aptitude area is this true. Taking into account that the A schools comprise 11.143subjects (a very large sample), the development of a new Administrative composite .would seem to beworthwhile, even if the Al. continue to be computed in the 'conventional. way.

Considering levels within aptitude areas, the Base groups (that group hi which the weights werederiied which were then applied to the target groups) would be expected to produce higher CI validitiesthan the Target groups because the equations that were instrumental in producing the K criterion werederiverin the Base grove. If the CI validities of the Base group are substantially higher than those of theTarget groups: moue benefit would be expected.from using the CI composite with schools at that level.However, thovidence argues the opposite case.

In the M area, the average AI-M validity for the Base schools is .45 and the average CI-M validity forthese schools is .54, an increase of 09. In the Taiget schools (see Table 4), the average AI-M validity is .41

'and the average Cl-M is .56, in increase of .15.

Table 4. Improvements in Prediction by C2 Composites,411,

Bale and Target Sehohls Compared

BaseSchools

0

AI (M) .45CI (M) .54

Difference

Ala) .35CI (A)

Difference 19

AE(G) AOCI (G) A2

Difference .02

Ai (E) .57'CI (E) .65

Difference .0f)

TomsSchools

.41

36

.15

.25

.55

.30

.52;55

.03

:56.66

.10

15

In the A area, the average Al-A validity is .35 for Base grwip schools and average Cl -A validity is .54.a difference of .19. In the A area Target schools, the average Al-A validity io.25 and average Cl -A validityis .55,

--G-area Base schools produce an average Al-C validity of .40 and average CI-G validity_ of .42, animprovelent of only .02. The Target schooll produce an Al-G validity of .52 and a Cl -C average validityof .55t an improvement of .03.

Overall, the average AI,E validity in the-Base groups is .57, which increases to .65 (a MBimprovement) when the CI-E composite is used. In the E area Target schools, the average AI-E validity is.56, compared with .'16 far the Cl -E composite. an improvement of .10.

In summary. the Cl composite produces more improvement in the Target schools in every instance(though the effect is small in the G area). This result is exactly the opposite of predictions. and the reasonthis effect should appear is unknown. At any rate, using the Cl compoeith in the Target schools (ratherthan the Al composites) would be more advantageous than using them in the Base schools.

There were very- large differences amoni individual schools in the amount of predictiveimprovement effected by the CI composite. These differences ranged from -.12 (school 23IX 1. G60) to+.53 (school 732X1. A60). There was no increase in predictive accuracy in only 12 of the 119tehools. Thevalidities of 45 schools improved at least .10 when the CI composite is substituted for the traditional Al

, composites, 24 schools improved at least .15. and 12 improved at least .20. Clearly, there are me,yindividual schools in which use of the CI composite could result in substantial improvement in predictive

_ .,efficiency.

IV. WACLUSIONS AND RECOMMIEPIDATIONS

The information contained in this report leads to the following conclusions and recommendatimis:

1. AcrossNall school*, the method of producing the Cl colvosite yields results t ubitantiatly betterthan the traditional method ("producing the Al composite. Thetionsl approach produces an average

i validity of .50, compared with .59 produced by the Cl "composites. The difference between the squares ofthese validity coefficients it .10 (35 minas 25), and the Proportional improvement (.10 .25) is .40.This Inst-number means-thit, starting with 25% predictive efficiency using the conventional Ale, a 40%improvement (raising 25% up toe%) can be made by forming the Ala in the manner described for the Cl

posses. if CI composites are used to select for some but not all the schools, machinate dramatic*ray be obtained (rig., 114X0, 531X4, 552X0, 566X1, 605X1, all the A schools, 362X4. andLasing Als computed in the traditional manner to select for some schools. and £1 composites

as in this study to select for others is not a serious problem. The only additional-procedurewould be tbecomputation And ret arding forksch enlistee of an additional set of compositesan'vial for a computer.

2. the Cl composites are net substantially improved by adding educational variables to the set of

& Attbough the primary objective of this study was not the evalhation of the predictive' efficiency ofconventional aptitude indexes, the Administrative AI, as currently constituted, shows up as such a

poor selector for sehodis in the Administrative area that this finding should be documented. Ascertainingthe Wicfity of this finding should be the objective of future research on ASVAB composites and if

efforts should be directed to the dv....ziopment of a new. Administrative Al to increase thealidity.of. this 'composite.

4. The K-composite procedure worked rather It is a procedure which should be mind not onlyin the context described herein, but also in academic prediction studies involving grade poikt averages offreshmen, sophomores, junior. and 'seniors collapsed into a single criterion group. The procedure couldalso he used in studies predicting rating criteria collected on the samelcale on subjects of different ranksand in other situations where criterion data are collected across groups of varying levels on scalesrestricted by'arbitrary upper or lower limits.

REFERENCES

Guilford, J.P. Fundamental statistics in psychology and education.-Newc: McGraw-Hill. 1950.

kieNentar, 1. Lost: Our intelligence? Why? Amaricifiisycho/ogist, 1964,, 19 (12), 871-882.

Wainer, H. Estimatiqf coefficients in linear models: It don't .nake nevermind: Psychological Bulletin,1976, 83, 21217.

Weeks, J.L., Mullins, Q., t# Vitola, B.M. Airman classification batteries )rom 1948 to 1975: reviewand evaluation. AlFHRL-TR-75-78, AD-X026 470. Lackland AFB, TX: Persc mel ResearchDivision, Air Force Human Resources Laboratory, December 1975.

N.

A

20i7