Medical careers begin as undifferentiated, and postgraduate training ends with most doctors specialised for a specific area of practice. Relatively little is known about the transition from the medical student, who can be seen as a relatively undifferentiated, totipotent 'stem doctor' 12, potentially capable of entering any speciality, through to the final, fully-differentiated specialist who is almost entirely restricted to one specialised area of medical work. Although medical career specialisation has been subject to a moderate amount of research (for reviews see e.g. 34), some of it going back over half a century (e.g. 5), much of that research has concentrated on the personal characteristics of individuals choosing particular careers (e.g. 678, on background factors in childhood influencing career choice (e.g. 8910), on associations with particular personality types (e.g. 11), on the careers of specific groups, such as women doctors (e.g. 12), on attitudes towards specific specialities, such as psychiatry (e.g. 1314) or anaesthetics (e.g. 1516), or has concentrated on the basic statistics necessary for workforce planning (e.g. 1718). There is, however, a lack of any broad theoretical framework in which to place career choice and specialisation.

UK medical education requires undergraduates to study a wide range of medical specialities, and most students will have sampled many of the broad areas of practice by the time they qualify. As a result, it is often assumed that students do not make their career choices until after they have finished at medical school, remaining agnostic about their final speciality choice until that time. However, not only medical school entrants (e.g. 19), but even medical school applicants, a year or so earlier, at the typical age of about seventeen, often have surprisingly strong preferences for, and particularly, against, some medical careers (e.g. 20). There is strong evidence, therefore, that career choice can be determined during or even before medical school (2122)). Thus, it makes sense to try and understand those preferences, which probably underpin eventual career choice.

Much research into medical careers does not take into account the broader research literature on non-medical careers (see 232425), or on socio-psychological models of the theoretical underpinnings of career choice (e.g. 26, 27, 2829). Consequently, medical careers research often fails to provide any broader theoretical framework or conceptualisation within which the empirical findings may be explained or which allow generalisations beyond the immediate data collected in the study (although there are exceptions, e.g. 3031).

The present study takes its origins in three separate sets of theoretical approaches, each of which examines different aspects of careers. None of these approaches, however, concerns medical careers specifically. Neither are they restricted to career choice in adulthood. Furthermore, at least one of them is specifically developmental, emphasising the processes by which career choice occurs and changes. The best place to begin this brief theoretical review is with the work of Gottfredson 27, who identifies the distinct processes of circumscription and compromise in career choice.

Careers differ in their demands, requiring different amounts of intellectual ability, manual skill, long-term commitment, or willingness to work in particular environments, and can be better suited to particular personalities, aptitudes, and physical dispositions. Individuals also differ, having different aptitudes, interests and abilities. Career choice therefore involves people considering the entire range of careers and then circumscribing those which they regard as broadly acceptable, making their eventual choices within that subset.

An important practical point highlighted by studies such as Gottfredson's is that choices tend to be negative, meaning that careers are rejected because they do not have attributes which are consonant with the person making the choice, rather than positively chosen for their special suitability.

Once circumscription has taken place, a number of possible careers still remain. The second stage of choice is compromise. Because of various practical constraints, certain careers are restricted in the number of people they can accommodate or they are unsuitable in other terms, such as their geographical location or the remuneration they can provide. The eventual career chosen is one that 'satisfices,' 32 being realistically good, though not optimal. The applications of this theory to medical careers are self-evident and describe many of the problems facing medical students and junior doctors.

Implicit in Gottfredson's conceptualisation is the concept of a map of careers. In her 1981 paper she provides an example a two-dimensional representation of 129 occupations which have been scored in terms of 'Prestige level' (high vs low) and 'sextype rating' (masculine vs feminine). When careers are mapped into this space, the process of circumscription involves drawing an area within which careers are acceptable to a person, being neither too masculine nor too feminine, nor being too high in terms of their prestige and hence effort required, nor too low, and hence insufficiently rewarding. A primary concern of the present study is the nature of the map underlying medical careers, and on which circumscription eventually takes place.

Perhaps the most influential study of the structure of career preferences is that of Holland 26, an overview and critical analysis of which can be found in the special issue of the Journal of Vocational Behavior published in 2000 (e.g. 25; see also 33 and 34). Holland's theory suggests that careers can be organised into six broad types, which can be represented around a hexagon (see figure 1), and which are often known by the acronym RIASEC, standing for Realistic, Investigative, Artistic, Social, Enterprising and Conventional. In Holland's original conceptualisation the specific orientation of the hexagon is arbitrary to rotation, but subsequent analyses have suggested that the hexagonal structure can be reduced to two dimensions 3536. One dimension runs from Realistic to Social, involving careers that are primarily Things-oriented rather than People-oriented. The second orthogonal dimension runs from midway between Enterprising and Conventional to midway between Artistic and Investigative, and involves careers varying from those that are primarily Data-oriented to those that are primarily Ideas-oriented. Holland's RIASEC model provides an appropriate two-dimensional space in which Gottfredson's circumscription model can apply 37.

Although Holland's work suggests how careers might be mapped, and Gottfredson's work suggests how career choices might take place within the space underlying those careers, a missing link in the overall picture concerns how individuals choose within the space. This is a significant question because individuals are expected to circumscribe in different ways according to their particular personalities and abilities. Ackerman 2829 has described how intellectual ability and personality relate to Holland's RIASEC model. Measures of intellectual ability primarily correlate with interest in the Realistic, Investigative and Artistic careers, people with higher verbal abilities preferring careers in Artistic and Investigative careers, and people with higher spatial and mathematical abilities preferring Realistic and Investigative careers. In contrast, measures of personality mainly correlate with the SEC components of RIASEC. Ackerman uses the Big Five typology of personality (see 38, 39), and shows that Extraversion primarily correlates with an interest in Social and Enterprising careers, whereas Conscientiousness correlates with an interest in Conventional and Enterprising careers. The personality dimension of Openness to Experience is to some extent a hybrid between intellectual ability and personality, and tends to correlate positively with Artistic, Investigative and Realistic careers, and negatively with Conventional careers. This pattern is similar to that which Zhang has reported in which the RIA cluster of careers relates to a deep approach to learning 4041, whereas the SEC cluster relates to a strategic approach to learning 42.

Between them, the models of Holland, Ackerman and Gottfredson provide, respectively, a good conceptualisation of i) the structure of careers and career preferences, ii) the correlations of careers with ability and personality, and iii) the developmental processes by which career choices are made. The question for medical education is the extent to which these approaches are appropriate for understanding medical career choice. If they are valid, then that will allow the much broader research literature from career choice in general to inform the more specific area of medical career choice. Underpinning the models of both Ackerman and Gottfredson is Holland's picture of a relatively simple, two-dimensional career map, onto which ability and personality can project, and on the basis of which career choices can develop. We therefore have two main objectives in this paper; firstly, to use data on career preferences from three separate cohorts of medical students, both at the time of application and in their final year at medical school, in order to derive a map of medical careers. And second, to assess the extent to which this specific map of medical careers is homologous to Holland's more general map of a broad range of careers.

The data collected in our studies consist of ratings of attractiveness of different medical careers on a five-point scale, ranging from 'Definite intention to go into this' through to 'Definite intention not to go into this'. However, our primary interest for the purpose of deriving a map of careers is not in career preference, but rather in career similarity. If a student has a preference for career A and career B, but has no interest in career C and D, it follows that career A is probably relatively close to career B on the map, and career C is relatively close to career D, whereas careers A and B are likely to be more distant from careers C and D. A matrix of similarities between all possible pairs of a large number of careers from a large number of students then allows one to construct the underlying map (just as, in a classic example, a knowledge of the geographical closeness, or the drive-time, between many pairs of towns in a country allows one to reconstruct a map of the country 4344). The statistical technique is known as multi-dimensional scaling (MDS).

Although conventional MDS can reconstruct the underlying map showing the relations between a number of objects, the map itself is arbitrary to rotation. Turning the map through any angle does not change any of the distances between pairs of objects, and therefore the axes of the map cannot be known – in the case of a geographical map, there is no indication of the north-south and east-west axes. The problem of the arbitrariness of dimensions can be circumvented by means of a variant of MDS known as INDSCAL (

In

D

Scal

Two different types of data have been used in the present study. The bulk of the analysis looks at data collected during studies of medical student selection and training, and can be used to map medical careers. A subsidiary, but important, analysis looks at a large convenience sample of people of different ages who were taking part in a survey about careers in general, and were asked about their interest in a range of careers, of which only a few were medical. These latter data allowed us both to calibrate specific medical careers in the context of the general Holland typology, and to validate the INDSCAL methodology for deriving a map of careers.

We consider firstly the general population sample since it both validates the method which we will subsequently use for the medical student samples, and also helps calibrate the axes.

The primary objectives of this study were to use the empirical method of individual differences scaling to derive maps of the underlying perceived structure of medical career specialities, and to assess the extent to which those maps are similar to those described by Holland in his hexagonal representation of the RIASEC groups of careers. That this method is a valid way of deriving Holland's structure in general is seen in figure 2, in which a broad range of non-medical careers is assessed by non-medical individuals, and the RIASEC structure is readily derived. Of particular importance is that because the analysis used INDSCAL, the dimensions are not arbitrary to rotation, and that the R-S dimension (corresponding to the Things-People dimension) and the IA-EC dimension (corresponding to the Ideas-Data dimension) are the basic underlying structure, as shown by Prediger 3553. The "Things-People" dimension also bears a strong similarity to the Technique orientation and People orientation which has also been described in relation to medical specialities 56.

The general population sample also rated four medical specialities, with Surgery and Anaesthetics at one extreme, and Psychiatry at the other, and these medical specialities differed principally along the R-S dimension. That Surgery and Anaesthetics are more concerned with Things, and Psychiatry is more concerned with People fits well with the reduction of Holland's hexagon to the two dimensions of Things-People and Ideas-Data. It is also worth noting that all of the four medical specialities are seen by the general public as being primarily concerned with Ideas rather than with Data, as surely befits medical careers.

The MDS analyses demonstrate that the representation of the various medical specialities by the medical student samples can be captured within a two-dimensional space, as Holland had suggested. The maps shown in figures 3 to 8 indicate that the structures are broadly similar across the three cohorts, and that although there are some minor differences between the applicants and the final-year students, it is the case that overall the similarities are more impressive than the differences. The crucial question therefore concerns whether the medical student maps are homologous to those of Holland's RIASEC typology. If there is a homology, then one may ask what are the Realistic, Investigative, Artistic, Social, Enterprising and Conventional specialities of medicine.

From scrutinising figures 3 to 8 we suggest that the RIASEC structure of medicine is typified by the six prototypical specialities of Surgery, Hospital Medicine, Psychiatry, Public Health, Administrative Medicine and Laboratory Medicine. It should be emphasised that in suggesting this we are not implying a direct comparability in the posts, rather a formal similarity within the limits imposed by being within the domain of medicine, as opposed to that of careers in general.

INDSCAL Individual differences scaling

MDS Multidimensional scaling

RIASEC Realistic-Investigative-Artistic-Social-Enterprising-Conventional

Speciality abbreviations in figures 3 to 8.

ADM Administrative Medicine

ANS Anaesthetics

ARM Armed Forces

BMS Basic Medical Sciences

DRM Dermatology

ENT Ear, Nose & Throat

FRN Forensic

GER Geriatrics

GPlrg GP Large Group practice

GPsml GP Small practice

GPsng GP Single handed

GUM Genito-urinary medicine

IND Industrial Medicine

INF Infectious diseases

LAB Laboratory (Haematology, Clinical Chemistry, etc.)

MED Internal Medicine

O&G Obstetrics & Gynaecology

O&T Orthopaedics & Trauma

OPH Ophthalmology

PED Paediatrics

PHM Pharmaceutical Medicine

PSY Psychiatry

PTH Pathology

PUB Public Health

RAD Radiology/ Radiotherapy

RDL Radiology

RDT Radiotherapy

RES Research

SRG Surgery

The authors declare that they have no competing interests.

ICM had collected the data in the various surveys over a number of years. ICM and KVP jointly decided how to do the statistical analysis, and KVP was responsible for the programming and data analysis. ICM wrote the first draft of the paper, which was revised by KVP, with both authors being responsible for the final draft.