PREDICTING HOLLAND OCCUPATIONAL CODES BY MEANS OF PAQ JOB DIMENSION SCORES

A sl udy wascunducl<'d on hul\' 10 obtain Holland'S codes for Soulh Afric~ 1I occupaliuns practic~ lI y ~nd emnom· ic" lIy by d~'ducing them from infurmation on the n~turc of the occupntiun (as d~>ri\'ed by means uf the Posi tion Anlllysis Questiunn;lirc). A discriminant analysis rcvealcd that un the basis of the J'AQ informnti"n tho.' occupationsmuld be distinguisho.'d clearly according tv tho:m~in orientations of their Amo.'rican codo.'s. Regression &.juations ,wre also dc\'~' lopcd to predict the mean s..'lf-Dirt.'Ctcd Search scorcs vf Ihe loccupntions on Ihe basis of their I'AQ inf'lrnMtiun .

Many ps},chologisIS, personnel practitioners, counsellors and advisers daily have to deal with problems s u rrou nding th e linking of an individual's characteristics to the requirements set by s pecific jobs.The available person al and occu patio n al information us ually h as to be in tegrated a nd interpreted by every investigaloror user w h o is interested in the results.Such interpretations are tedious and t ime con suming and usu a ll y occur in a fairly subje<:tive m a n ne r.The real problem w ith regard to voca tiona l counselling a nd career planning, and this also applies in the case of the selection a nd placement of in• dividuals, is therefore to collale meaningfully or 10 integrate the two sets of information when they are available (Smith, 19'75;Sparrow, Patrick, Spurgeo n & Barwell, 1982).Holland (1966) developed a theory in an attempt to organize and interpret the availa ble information on occupational behaviour.This ap proach has ulility value as it o ffers a possible link between information about the individual and the world of work.According to Holland (1'l73) h is theory o f occupational choice is based on t he premise that ocr;upa tional interests are rega rded as o ne of the aspects of person ality.Consequently an indication of a person's occupational interests ca n also be regarded as a parlial expression o f h is personality.
The essence of Holland's theory is his categorization o f individuals into six particular personality types and the associatio n of Ihese ty pes with six corresponding wo rk environments (H olland , 1987).Holland states that, generally s peaking.every person corresponds with o n e of the personality types, but is also influe nced by a second or even a th ird type, all of which contribu te to the person's ability to deal with his environment.The s ix main personality types of Holland are: Realistic (R); Investigalive (I); Artist ic (A); Social (S); Enterprising (E); a nd Convention al (C).
Holland's theory thus p resents a method in which personal and occupatio nal in forma tio n ca n be utilised effectively by Requests for reprints should be add ressed 10 R. I' .van der Merwe, 386 Cape Road, Ferng!cn, Port Elizabet h 6().1S.
Zl linking a person's code to an occupalion that has a corresponding code.Holland's SeIf• D irected Search (50S) (1981,1985) is therefore o ne of the very few occu pational choice q u estionnaires that is based on a theoretical framework involving both Ihe individu al and the world of work.His theory e njoys growing statu s and esleem in th e United States of America and h is SOS is one of Ihe m ost generally used psychometric instrumenls in occupation al ch oice.
In order to dete rmine the code of a particular occupation, var• io us methods can be u sed .According to one of the methods, large groups of practitioners of t he occupation in question comple te the SDS after which the average code fo r that occupation is dete rmined on the bas is of the re levant resu lts.Another popular method is to use job analysis data and to assign part icu lar codes to the occu pation on the basis of specif• ic job conte nts.Experts ca n then be used to estimate the code of an occupation o n the b.. l s is of the particular job conle nts.Instead of experts making an estimate of an occupatio n's code, the code can a lso be dete rmined by developing a model on the bas is of Ihe obtained job analysis information.Wh en developing s uch a model, it is advisable that informa• tion o n the jobs should be obtained in a scienlifk manner, The Posilion Analysis Ques tionn aire (PAQ) is well s uited for this purpose.II is a slructu red q u estionnaire by means of which jobs are a nalysed a nd described as the work relati ng to them is carried out in practice.The question naire is an ex• ample o f behaviour description classification that places the e mphas is m a inly on w h at the wo rkers have to do in o rder to complete the ir work (McCormick,1<714;Palmer & McCormick,1%1).In terms of this approach not only can separate be• havioural areas be identified in respect of a large variety of jobs, but each behavioural area can also be classified into smaller general wo rk elemenls.With every work element assessed o n a points scale, the PAQ system makes it possible 10 desc ri be jobs in m eaningful and qua ntifiable units of work information.Work d ata in this form m eet the slricl requireme n ts set wilh a view 10 resea rch, namely that it sh ould be possible to q u an tify o r categorize the relevant variables reliably (McCormick, Cunningham & Gordon , 1%7;Smith, 1981).
As the n alure of occu pations and incumbents ca n differ from one cou ntry to the nex!, it mean s that w h en Ho lland's approach is used in South Africa, the q u estion arises as to the rclcvanceof the American occupational codes to South African conditions.Determining codes for occupations in South Africa by obtaining 50S profiles of incumbents would be an onerous and expensive task.The use of experts to deduce the codes from job descriptions is convenient but there may be dangers with regard to subjectivity.The establishment of an objective method to assign Holland occupational codes to Sou th African occu pations can make a very useful contribution to the fi eld of occupational counselling and career planning, and to the selection and placement of individuals -the reason being thai the link between a person and an occupational field would be established more easily and can take place more scientifically.

MFfHOD
In the present investigation it was attempted to develop a model by means of w hich the Holland occupational codes, as obtained on the basis of the mean SDS scores of incumbents, can be predicted by using job description information as obtained by means of the PAQ.
Test group Sixty occupations were identified on the basis of the Holland model and it was attempted to involve approximately te n incumbents per occupation.These incumbents were drawn from large organizations and companies.For the sake of homogeneity it was attempted to include preferably only white men in the sample.They had to have a mi nimum of two years practical experience in their particular field of work and an effort was made to keep the total range of ages within each oc<:upation as limited as possible.job satisfaction was also borne in mind and on the basis of results obtained by means of the Job Satisfaction Index 051) the final decision was taken as to whether a person could be included in the sample or not.The final sample group co nsisted of 576 subjects.

MEASURING INSfRUMENTS
Self-Directed Search (50S) The SDS is Holland's occupational choice questionnaire fo r gathering specific information on a person .It is comprehensive and evaluates a person's act ivities (what he likes to do), his competencies (i.e. the present competence of an individual based upon the experience he gained in the past), his selfassessing abilities (assesses himself with regard to his abilities and skills in comparison with other persons in the same age group), as well as his occupational likes and dislikes.
The questionnaire consists of 228 items and on the basis of the individual's responses a total score is calculated for each of the six personality dimensions.A combination of the individual's th ree highest personality dimensions, giving an indication of his experience and occupational preferences, is used to determine his three-leiter (main class and two subclasses) SDS code (personal code).This SDS code is composed of any combination of three of Holland's six personality dimensions: R, I, A, S, E, C.
Position Analysis Questionnaire {PAQI The fact that the PAQ is objective and does not discriminate s ubjectively between jobs makes it, from a scientific point of view, a very suitable instrument.It can be used very s uccessfully in a large number of situations in which, on the one hand, the characteristics of jobs are compared with one another, or, on the other hand , the characteristics of jobs are compared with aptitudes of persons.It can also be used to determine the aptitude requirements set by a particular job for its incumbent.Although the PAQ has many possible uses, onl y its job analysis aspects are often used as it is a good system for identifying the structure of human work, quantifying it and particularly for determining aptitude requirements for different jobs and groupings (Carter  The results of the 194 items of the PAQ are reflected in 32 specific (divisional) and 13 general (overall) work dimensions.This means that a score is obtained on each of these 45 d imensions for every job.By using these scores in certain regression equations, an indication can be obtained of the human abilities that are required for funct ioning in a specific job (McCormick, DeNisi &. Shaw, 1978;McCormick, Mecham &. jeanneret, Ima, 1977b, 1977c, 1977d).The expected human traits or characteristics such as interest, personality and aptitude are therefore deduced from the job description information.Average scores on the subtests of the General Aptitude Test Battery (GATB) can also be predicted.
Job Satisfaction Index OSI) Brayfield and Rothe (1951) used Thurstone and Likert's scaling method in the construction of their Js l.From an original 246 items they designed a final scale consisting of 18 items.Mauer (1976) administered the Js l to mine workers and subjected the 18 items to item analysis.On the basis of the resul ts he shortened the scale to 16 items.
From published information it is evident that the jSl has satisfactory reliability and validity; moreover the questionnaire is short and administering it takes up little time.For these reasons it was regarded as a s uitable instrument for determining the job satisfaction of the s ubjects to decide whether they could be involved in this investigation or not.A cut-off point of 48 was decided on, which was the sum of the mean values of the individual items.In terms of this some 10% of the persons d id not meet the set requirements and were not included in the investigation group.

Statistical procedures
The final test group consisted of 576 s ubjects, each of whom completed the SDS and the lSI.In the case of the sixty chosen occupations a minimum of two, and in some cases even three, PAQ analyses were carried out for each oc<:upation.Following the testing, scoring and processing of the data an average score for each of the sixty occupations on each of the six SDS fields was available -as calculated from the obtained sDs results of the 576 subjects.On the other hand, for each of the sixty occupations there was also a score on each of the 13 overall dimensions of the PAQ -as obtained from the results of the PAQ analyses that had been carried out.
In the present study it was firstly attempted to evaluate the applicability of Holland occupational codes to South African oc<:upations.The similarities were checked visually and t-tests were also u sed .Subsequently it was attempted to determine w hether the occupations could be classified on the basis of their PAQ dimension scores in the six previously determined main occupational groups of Holland .For this purpose a discriminant analysis was carried out on the basis of the observed job dimension scores for every occupation.UlStly the linear relationship between the two sets of data were investigated by means of Pearson correlation coefficients and multiple regression equations to determine whether the PAQ dimension scores could be used to predict scores on the SDS personality dimensions.

RES ULTS
Firstly it was ascertained to w hat extent the empirically determined occupational codes of the sixty South African occupations concerned corres ponded with the Holland occupational codes that had originally been allocated in the United States of America.
When the sequence of the three letters in an occupational code was ignored, the original American Holland oc<:upational codes and the South African empirically determined occupational codes were the same with regard to one letter in 60 (100%) of the cases; wi th regard to two letters in 52 (86,7%) of the cases; and with rega rd to all three letters in 24 (40%) of the cases.
When the sequence of the three letters in an occupational code was taken into consideration, the original American Holland occupational codes and the South African empirically determined occupational codes corresponded w ith regard to their first letters in 34 (56,7%) of the cases; w ith regard to their first two letters in 11 (18,3%) of the cases; and with regard to all three of their letters in 7 (11,7%) of cases.Better results were therefore obtained by ignoring the sequence of the letters in an occupational code than by taking them into account.
Subsequently it was determined whether the incumbents could be classified on the basis of their mean e:npirically observed SDS scores in the thirty previously determined suboccupational groups (two letter codes according to the Holland model).As no purely AC, AR or CA occupations could be found in the case of the present sample, only Tl of the actual 30 suboccupational groups were re presented in this case.
When the sequence of the letters in a two-letter code was ignored the original American suboccupational codes were the same with regard to one letter in 26 (%,3%) of the cases and with regard to both letters in 9 (33,3%) of the cases.
When the sequence of the two letters in a suboccupational code was taken into consideration, the original American suboccupational codes and the South African empirically determined s uboccupational codes corresponded in respect of their first letters in 18 (66,7%) of the cases and in respect of both their letters in 6 (22,2%) of the cases.Better results were therefore obtained by ignoring the sequence of the letters in a suboccupational code than by taking them into account.
It was also determined whether the incumbents could be classified on the basis of their empirically observed SDS scores in the six previously determined main occupational groups of Holland (R, I, A, S, E, C).With regard to each main occupational group Significant differences were found, by means of t-tests, between the calculated means of the incumbents of jobs who had been classified into that occupational group and the incumbents who had not been classified inlo that specific occupational group.Particularly good results were therefore obtained by attempting to distinguish the six occupational groups on the basis of the incumbents' empirically observed SDS scores.These results appear in Table 1.
Subsequently it was determ ined by means of discriminant analysis whether the occupations could be classified on the basis of every occupation's obtained PAQ dimension scores in the six previously determined main occupational groups (R, I, A, S, E. C).
[n order to bring the number of variables (13) into a better ratio to the number of observations (10) for each main occupational group, with a view to carrying out a discriminant analysis, it was de<:ided to leave out job dimensions that had the lowest d iscriminatory val ue between occupational groups.Analysis of variance (ANOVA) was subsequently carried out to obtain information on the discrimination value of each of the 13 job dimensions, On the basis of these results it was de<:ided to make use of the seven job dimensions that could at least at the 10% level identify significant differences between the six main occupational groups in further processings.
The data should meet certain requirements before a multivar• iate discrim inant analysis can be carried out (Betz, 1987;Du Toit & Stumpf, 1982).The requirements in question are the following: -The data set should be derived from a multivariate normal population; -with equal subgroup covariance matrices; and -the s ubgroups should be a collections of independent data sets.
With regard to the present data set it is accepted that it originates in a multivariate normal population although the subgroups contain fewe r than thirty observations -as is al• ways the case for the central limil theorem.The subgroups are also independent as presupposed by premise (iii), as ev• ery observation is limited to one particular group only.In order to check the equality of the covariance matrices, a chi-square test was carried out.This yielded a value of 242,73 with 140 degrees of freedom (p < 0.01).As the chi-square value is significant, a linear d iscriminant analysis was not desirable (Thabachnick & Fidell, 1983) and consequently a quadratic d iscriminant analysiS was carried out that made use of separate covariance matrices.The res ults of the classification made on the basis of this analysis are shown in Table 2.The results in Table 2 show clearly that the poorest classification occurred in the case of the Artist ic group, in which only seven of the ten posts (70%) had been placed correctly according to their job dimension scores.P.lrticularly good results .. ..-ere therefore obtained by classifying occupations on the basis of their PAQ dimension scores according to Holland occupational codes and the conclusion can be reached that this is data in respect of six clearl y different occupational groups.In order to dcvelop a model for predicting the Holland occup.ltionalcodes by mc,ms of job descriptio n information, the linea r rela tionship lX"tw('cn the obtained PAQ dimens ion scores o( an occup,ltio n an d the mean s of each of the six 5 0 S personalit y dimensions for that occupation had to be checked .In order to invcstigate thl' rel,ltillnsh ip bt'lw('('n Ihe obl.linedPAQ dimension SCO Tl'S and 1h(' tllt'iHl S of (" leh of the six 5 0 S personalit y dimensions of an occup.llion,1\'o.1 T50" correlation coefficients were calculdt('d.A Sh: pw is.• multiple n>grcssion analysis W, l Sl"onductcd with the m e,ln scores on each of the six 50 S personality d imensions as de pende nt v.uiables ,md the 13 overall PAQ dimension scores .15independent variables so that tht' va ri abll's that plaYl'd the most useful role in the predictio n could be determined .This W;IS d one by us ing the F-test to investigate the signifi~ancc of the incrcaSt' in v.u-ianCl' from the first h) the last step of the model (I't-dhaz ur, 1982).
If an increase in variance betwec n th(' first and last ste p of the modd is found wh ich is s ignificant at l('ast at the 5% level of confidence, o ne continues by comparing the fo llowing equation (Step 2) wi th the last step.This procedure is repeated fo r the s ubSl..' <luent s teps until that specific s\{'p (regression equation) is found whoS(' variance dl)('s not differ s ignificantly from the last equation.By following this appro..lch a specific regress io n equatio n is therefore identifie d as th{' cut-off point for each of the s ix 5 DS personality dime nsions (secTable3) .T he specific reg ression ('(Iuations as obtained at every p;lrticular cut-off point (ya + b ,X, + b 2 X 2 + b,X 1 + b~X~ + b~X~) arc repo rt('d in Table 4. 5 (1,5503 "13,22"   Conn'nli"Il.)! 0,57t19 5 05322 12,29- ' p < 0,00 1 By using the empirically obtained PAQ dim{'nsions of each occupatio n and these calculated regress ion equatio ns in resp{'(t of {'Very cccupatio nal group, predicted S OS scores for each of the s ixty occu patio ns were calculated and the S DS code concern{'d was deter mined o n the basis of the threi!h ig hest scores.Subsequently the s imilari ties that occu rred between the empiricall y d ete rmined occupational codes and the codes that weT(' o btained by means of the regreSSion equations were inves tig;lted. Table 5 shows a sum mary of the s imilari ties that occur betwC{'n these two sets of data when the sequ{'nce of the three le ll{'rs in iln occupational code is ig nored .was ig no red, the South African empirically dete rmined occupational codes a nd the occupatio nal codes that were obtained by means of the reg~ss ion equations were the same with reg.1Td to one letter in 60 (100%) of thl' cases; wi th regard to two leiters in 54 (90%) of th .... cases; a nd w ith rega rd to all three letters in 28 (46,7%) of the Cilses.
Table 6 shows a sum mill)' of the s imila ri ties thai o.x ur be\w('en thl'Sl' twu sets of d.lta when th(' seque nce of the t hree Ielt("'rs in o,:cupatio nil l cod{'s was til ken into consideration and consequently had to Ixthe sa ine.

DISCUSSION
T he differences betwecn the empirically determined occupational codes and the original American Holland occupational codes of the s ixty occupations can be ascribed to various causes.It should be remembered that the American Holland occupatio nal codes do not necessarily apply unchanged to South African occupations.Empirically the South African code for an occupation can therefore differ from the overseas code.In addition the number of persons per occu pation who were involved in this study was s mall (about ten) and this too, could have influenced the occupations' mean SOS scores, on which the e mpirical occupational code is based .Theconclusian can therefore be reached that care should be taken with the use of American Holland occupational codes in the South African context as it would seem that unchanged they are not necessarily applicable to South African conditions.
By using suboccupation groups (two-letter codes), better results were obtained than in the case of the first-mentioned results where the individual occu pational codes were CO Illpared wit h on(' <lnother.In the case of the s uboccu pation groups the differences that occur between the two sets of data can be ascribed to mainly the same reasons as in the case of individual occupations.HowC\'er, it appears that according to the present resul ts the d ifferent s uboccupation groups can to some extent be classified on the basis of the mean empirically observed SDS scores of the incumbents.Consequently these findings tend to support the Holland model and its applicability to South African occu pations.Care should neverthdess be taken w ith the usc of American Holland occupational codes in the South African context as it would seem that unchanged they do not necessarily apply to Sou th African conditions. Subsequently it was ascertained Ihal Ihe incumbents in the six previously determined main occupational groups (according to the Holland model) can be classified on the basis of Iheir empiricallyobscrvcd SDS scores and that th is was data on six clearly different occu pational groups.These findings lend positive support for thc Holland model and its applicability to South African occupations.
The Sixty occupations involved in this investigation could 10 a great extent be classified into Ihe six previously determined main occupation.llgroups of Holland if only each occupation's obtained PAQ dimension scores were used.By therefore using job analysis information instead of occuptional or personal information, the six different occupational groups of Holland could still be classified and the conclusion can be reached that this is data on six clearly different occupational groups.These findings le nd fu rt her positive s upport to the Holland model and its applicability to South African occupations.They also illustrate how useful the PAQ is in making a clear distinction between different occupational groups.Although incumbents in Ihe six previously d etermined main occupational groups (according to the Holland model) could be classified o n the basis of their empirically observed SOS scores, it does appear from the results of this investig.ltionthat care should be taken wi th the use of American Holland occupational codes in the South African context as it \'IIOuld seem that, unchanged, they do nOi necessarily apply to South African conditions.
By means of regression equations (as developed from the empirically determined SDS codes) and the obtained PAQ d imension scores, SOS codes were developed that generally compare very well with the empirically determined SDS codes.Subsequently a relationsh ip exists between the obtained PAQ dimension scores of an occupation and the means of each of the six SDS personality dimensions for that occupation .PAQ personality dimensions can therefore be used to draw conclusions on the desired personality dimensions in the case of a specific occupation .
From the above it can be concluded that job dimension scores may be used to predict personality d imensions and this link between information about the world of work and the individual has far-reach ing implications in practice.The utility value in the case of personnel selection and placement, career planning and occupational guidance is that the link between a person and an occupational fie ld can be made more easily and also more scientifically.
The regression equations that were compiled, should be extended in the course of time w hen details concerning a greater number of ;obs and incumbents become available.However, the present results are so positive that they can be regarded as useful for the praxis.Such comparisons undoubtedly have w ide application possibilities in various fields and can greatly increase the scientific nature of voc.1tionai counselling and personnel selection.
of the three leiters in an occupational codl'