1. Top of page
  2. Abstract


To describe motivations correlating with subspecialty choices, particularly rheumatology.


A total of 179 respondents answered queries about various aspects affecting specialty and subspecialty choice with ordinal ratings of importance. Likert scale response data were analyzed to determine independent predictors of being a rheumatology fellow. Multivariate logistic regression analyses were used to develop models predicting rheumatology fellowship. Factor analysis methods to condense the individual responses into fewer underlying variables or factors were employed.


While every group ranked intellectual interest as more important than all other responses, its score in the rheumatology fellow group was significantly higher than that in the medical student group. A model using 4 composite variables based on prior literature did not fit well. Exploratory factor analysis identified 5 underlying motivations, which were designated as time, money, external constraints, practice content, and academics. All motivations except money were statistically significant, with the rheumatology fellow group attributing greater importance than medical students to time, practice content, and academics, and lesser importance than medical students to external constraints.


Values and motivations leading toward rheumatology subspecialty choice can be traced to identifiable factors. Intellectual interest appears to be split between 2 distinct significant variables: practice content and academics. Time or controllable lifestyle, external constraints, practice content, and academic issues appear to be important influences on the choice of rheumatology fellowship. Such variables appear to reflect underlying values and motivations.


  1. Top of page
  2. Abstract

Current economic conditions suggest that future demand for rheumatology services will increase, yet will have to be met by fewer and more flexible rheumatologists within constrained resources using practice redesign and innovation (1–5). Rheumatology may also meet increased needs through restricting the access of some patient populations for whom rheumatology consultation is less critical in light of disease-modifying therapy (4), and by increasing the use of rheumatology extenders (3). One additional practice redesign suggestion is to uncouple patient visits for laboratory or rheumatology extenders from those to rheumatologists (6–8). Despite practice redesign efforts, however, matching supply to demand will be difficult (9).

Recruiting, in this professional context, is really “matching” or “career counseling” so that a resident's values and motivations match well with the subspecialty choice, as well as contribute to higher career satisfaction. This study aimed to describe what professional and personal characteristics distinguish rheumatology fellows from younger trainees; the intention was to add to the understanding of values and motivations, and thereby to support future rheumatology recruitment. A conceptual approach to empirically derive grouping from data was employed. If the responses attaching importance to each query reflect underlying motivations or values, the response to one item could be correlated, according to these underlying values and motivations, with the response to others. Exploratory factor analysis represents a method without a priori assumptions about grouping. This analysis aims to uncover the underlying structure of a set of variables and reduce many variables to fewer that are more amenable to statistical and conceptual analyses (i.e., to reduce dimensionality). One might consider the analogy of a parent recognizing her child from the outline of bulges moving under a blanket; in this context, what one sees are the response variables, and what the parent infers represents an explanation of which response variables are correlated and why (i.e., the underlying or latent factors).


  1. Top of page
  2. Abstract

Survey respondents.

A total of 179 individuals at 4 levels of medical education were surveyed. The rheumatology fellows group (n = 75) included an ad hoc group of adult rheumatology fellows (n = 20), who responded on paper questionnaires at the 2005 American College of Rheumatology (ACR) Annual Scientific Meeting, a 25% sample group (n = 46) of first-year US adult rheumatology fellows, who completed the survey in 2006 via the Internet, and a sample group (n = 9) of pediatric rheumatology fellows who were folded into the rheumatology fellows group. The control group comprised lower-level trainees at 1 institution (Virginia Commonwealth University [VCU]): general internal medicine residents (n = 25), third-year medical students before or after medicine clerkship (n = 48), and faculty and other subspecialty fellows not in rheumatology (n = 31). Demographics of the survey participants are shown in Table 1. Due to the small number of nonrheumatology fellows, particularly procedural-based fellows such as those in cardiology and pulmonary medicine, the analysis concentrated on the contrast between medical students and rheumatology fellows.

Table 1. Demographics of survey participants*
ParticipantsNo.Male, %Married, %With children, %
  • *

    MS = medical students; IM = internal medicine.

  • Not rheumatology.

Total participants179594937
 Rheumatology fellow66646771
 Pediatric rheumatology fellow90560
 MS prior to IM clerkship20552515
 MS after IM clerkship2868180
 IM resident25643612
 Other subspecialty fellow26586538
Marital status
 No response1   
Children, no.

Survey development.

In this study, we focused on factors influencing subspecialty choice, especially rheumatology. First, an initial 42-item questionnaire was developed from a questionnaire validated in previous research, contrasting those persons selecting primary care careers with those who did not in the 1980s and 1990s (10). A total of 16 items were reconstructed from the published articles of Schwartz et al (10, 11), and these were supplemented by a phone interview with the investigators. Next, focus groups of medical students, internal medicine residents, fellows, and faculty were used to develop additional items about “controllable lifestyle” and time control (as delineated in the literature beginning in 1997). The final survey queries elicited 29 responses relating to specialty and subspecialty choice, specifically, demographics, control of time and work hours, income, and job opportunities. The VCU Institutional Review Board approved the questionnaire.

Many items were similar in content to the previously published sections (4, 5, and 6) in the questionnaire used by the ACR Training and Workforce Committee, Subcommittee on Medical Student and Resident Recruitment (12). The participants were asked to respond to items using a 5-point Likert scale (where 1 = not important at all, 2 = somewhat important, 3 = important, 4 = very important, and 5 = critically important) in order to rate the importance of a factor on specialty choice. Additional answers to questions about the most desirable and undesirable features of the respondent's current job or level of postgraduate year (PGY) were collected as free text, categorized post hoc, and assigned dummy numerical values. For example, in Table 2 one post hoc characterization is listed as “hate long hours.”

Table 2. Univariate comparisons among surveyed groups*
 Rheumatology fellowMedical studentInternal medicine residentOther fellow
  • *

    Values are the mean from each group. Rheumatology fellows group comparisons tested by nonparametric analyses followed by correction for multiple comparisons.

  • Significantly different from rheumatology fellows group.

  • Mean score for literature-derived factors based on a priori grouping. None differed between rheumatology fellows group and any other.

 Likelihood of litigation2.362.802.482.48
 Job opportunity3.353.163.483.52
 Time control3.793.363.523.00
 Number of hours3.673.263.483.00
 Call in practice3.373.003.042.90
 Spouse career2.282.812.202.14
 Hate long hours0.170.400.560.38
 Patient type3.783.133.403.52
 Community service2.703.042.642.52
 Social need3.073.002.603.00
 Continuity of care3.583.003.482.67
 Training length2.562.662.922.19
 Training intensity2.792.983.002.90
 Difficulty obtaining training position1.922.722.281.95
 Intellectual interest4.473.934.083.90
 Nonfaculty role model2.312.532.422.60
 Ability to fix problem2.783.333.203.38
 Academic medicine2.862.612.402.19
 Urban practice2.932.892.322.45
 Match skill3.463.493.243.57
 Own age2.102.341.962.00
 Life experience2.082.832.002.24
 Time off3.783.653.363.43
 Clinical role model3.072.892.843.00
Literature factor model
 Time off17.0416.5615.9214.81
 Patient altruism13.1412.2312.1211.71

Statistical analysis.

This study reports differences between rheumatology fellows and younger trainees. The first analyses were univariate, using Wilcoxon's nonparametric comparisons, followed by false discovery rate correction for multiple comparisons (13) (Table 2). Then multivariate analyses were done with multiple logistic regressions using various models (Tables 3 and 4). Status as a rheumatology fellow was the dependent variable, which was scored as 1 for true or 0 for false. The initial models used independent variables derived from responses to individual queries, with the set of independent variables chosen from literature-based responses or from literature plus exploratory responses. The literature model used the first 17 response variables (“likelihood of litigation” through “difficulty in obtaining residency position”), and the shown significant odds ratios (ORs) represent the likelihood of rheumatology fellowship for each increment of 1.0 along the Likert response scale of 1–5.

Table 3. Comparison among explanatory models comparing rheumatology fellows with medical students*
 Literature modelLiterature plus exploratory modelLiterature factor model
  • *

    Values are the odds ratio (OR) unless otherwise indicated. AIC = Akaike's information criterion.

  • Responses represented in full model as outlined. Significant by stepwise selection. ORs represent likelihood of rheumatology fellowship for each increment of 1.0 on the Likert response scale of 1–5.

  • Factors represent summed responses as grouped by pay, time off, patient altruism, and training. Factor score is the strength of the underlying factor, not the Likert 1–5 score of an individual questionnaire item.

  • ORs represent likelihood of rheumatology fellowship for each increment of 1.0 on the Likert response scale of 1–5.

  • §

    Significant by stepwise selection.

Full model, AIC111.167107.019139.555
Reduced model, AIC111.87194.809140.445
 Factor: pay  0.889
  Likelihood of litigation   
  Job opportunity   
 Factor: time off  1.160§
  Time control2.015  
  Number of hours 2.925 
  Call in practice   
  Spouse career   
  Hate long hours   
 Factor: patient altruism  1.147
  Patient type   
  Community service0.4850.425 
  Social needs   
  Continuity of care2.9342.922 
 Factor: training  0.728§
  Training length   
  Training intensity   
  Difficulty obtaining training position0.2980.152 
 Prestige 2.133 
 Intellectual interest   
 Nonfaculty role model   
 Like fixing problems   
 Academic medicine 2.143 
 Urban practice   
 Matching skills   
 Own age   
 Life experience   
 Time off   
Clinical role model   
Table 4. Comparison among explanatory models*
Response itemsFactors 
TimeMoneyExternal constraintsPractice contentAcademicsOR
  • *

    Akaike's information criterion 99.606 for full model and 97.856 for reduced model. Factors 1–5 used to derive factor scores, then in logistic regression analyses. OR = odds ratio (favoring rheumatology fellowship for each 1.0 increment of factor score; cannot be directly related to each response).

  • Significant.

  • Correlations (responses with factors) >0.4 deemed significant.

Factor 1: time     2.081
 Time control0.870. 
 Number of hours0.810. 
 Call in practice0.750.170.110.02−0.06 
 Time off0.680. 
Factor 2: money     1.161
 Likelihood of litigation0.240.540.170.01−0.18 
 Job opportunity0.210.640.310.150.24 
 Training length0.360.440.300.12−0.20 
 Nonfaculty role model0.110.430.300.10−0.28 
 Clinical role model0.000.780.080.150.06 
Factor 3: external constraints     0.248
 Spouse career0.300.150.500.13−0.13 
 Difficulty obtaining training position0.010.230.600.10−0.04 
 Own age0.280.110.55−0.070.01 
 Life experience−−0.14 
Factor 4: practice content     2.042
 Patient type0.110.02−0.040.73−0.03 
 Continuity of care0.11− 
 Intellectual interest−0.010.06−0.120.410.44 
Factor 5: academics     4.300
 Intellectual interest−0.010.06−0.120.410.44 
 Academic medicine−0.320. 
Hate long hours−−0.27 
Community service0.−0.48 
Social needs0.210.330.150.38−0.09 
Training intensity0.−0.14 
Like fixing problems−0.180.300.360.17−0.31 
Urban practice− 
Matching skills0.100.300.180.360.10 

The literature plus exploratory model used the same 17 response variables plus an additional 12 response variables (“prestige” through “clinical role model”), and the shown significant ORs represent the likelihood of rheumatology fellowship for each increment of 1.0 along the Likert response scale of 1–5.

Alternatively, responses were assembled into factors representing composite scores by literature-based grouping (10–12) of individual response data and then used as independent variables for multiple logistic regression, i.e., the literature factor model. The literature factor model included 4 response variables consisting of the sum of individual responses: 1) pay = likelihood of litigation + job opportunity + remuneration + loan, 2) time = time control + number of hours + call in practice + spouse career + family + hate long hours, 3) training issues = training length + training intensity + difficulty obtaining training position, and 4) altruism = patient type + community service + social needs + continuity of care. Therefore, values could range from 0–20, 0–30, 0–30, and 0–15, respectively. The literature-based grouping model of specialty choice is schematically shown in Figure 1.

thumbnail image

Figure 1. Literature factor model (prestudy hypothesis) schematically shown as a scale-free node (hub) model affecting specialty choice.

Download figure to PowerPoint

Finally, exploratory factor analysis was used to define numbers of factors, the responses correlated with each factor, and factor scores. Multiple logistic regression was used to examine factor scores used as independent variables (factor score model), which is illustrated in Figure 2. The literature model was derived from prior literature about student and resident motivations regarding primary care (10, 11) and 1 prior study about rheumatology (12). The model tested used all the responses for the items derived from literature as represented in Table 3 (literature model), and Akaike's information criterion (AIC; a parameter estimating model fit, with the lowest value representing best fit) was 111.167 for the full model. Using stepwise selection to generate a reduced model, the fit did not improve (AIC 111.871) with 4 statistically significant independent variables: the responses for time control (OR 2.015, P = 0.0083), community service (OR 0.485, P = 0.0130), continuity of care (OR 2.934, P = 0.0005), and difficulty in obtaining training position (OR 0.298, P < 0.0001). This would suggest that compared with the medical student group, rheumatology fellows assign greater importance to time control and continuity of care and lesser importance to community service and difficulty in obtaining training position.

thumbnail image

Figure 2. Factor score model schematically shown as a scale-free node (hub) model of subspecialty choice informed by factor analysis. hrs = hours.

Download figure to PowerPoint

Next, with the addition of other response items representing exploratory variables to the original literature-derived items (literature plus exploratory model, Table 3), the model fit was improved substantially (AIC 107.019). Stepwise selection found even better fit (AIC 94.809), with 3 of the 4 original items (not time control) retained: community service (OR 0.425, P = 0.00076), continuity of care (OR 2.922, P = 0.0024), and difficulty in obtaining training position (OR 0.152, P < 0.0001), along with an additional 3 items, which were number of hours (OR 2.925, P = 0.0015), prestige (OR 2.133, P = 0.0234), and academic medicine (OR 2.143, P = 0.0101). Therefore, one could infer that the rheumatology fellow group attaches more importance to number of hours, prestige, and academic medicine than does the medical student group. A literature-derived factor model (Table 4 and Figure 1) based on a priori grouping can be tested by deriving 4 variables represented by composite scores of the individual questionnaire items, representing pay, time off, altruism and patient orientation, and training issues.

Finally, we used factor analysis methods to gather and weight correlated responses. The first question is to determine how many factors are present. The first 22 factors accounted for 90% of variance, 10 factors had eigenvalues above the mean of 1.138, 12 had eigenvalues >1, and the scree plot showed leveling off after 5 factors. The cumulative proportion of variation after 5 factors was 0.8277. We therefore chose 5 as the number of factors. Next, we used a rotation method to help simplify interpretation (varimax rotation, which maximized variances of square loading for each factor). All analyses were done with Statistical Analysis System, version 9.2, for Windows (SAS), with the single exception of an Excel worksheet (Microsoft) for false discovery rate.


  1. Top of page
  2. Abstract

We first examined the responses among all subjects for multivariate normality; because we found kurtosis and negative skewing in the “love money” characterization, we excluded it from analyses. Table 2 shows mean scores of responses from the rheumatology fellow group compared with those from other groups. Compared with the medical student group, the rheumatology fellow group attached lower importance scores to hate long hours, difficulty in obtaining training position, fixing problems, and life experiences, and higher importance scores to patient type, continuity of care, and intellectual interest. Compared with smaller samples in other comparison groups, the rheumatology fellow group attached significantly less importance to hate long hours than did internal medicine residents (possibly reflecting the demands of their current jobs) and significantly more importance to continuity of care than other subspecialty fellows. Each group ranked intellectual interest higher than any other response from the group, but the numerical importance rating was higher in the rheumatology fellow group than the medical student group.

We next examined models that compared rheumatology fellows with younger trainees, specifically the medical students (Tables 3 and 4). Univariate comparisons of the composite scores were not significantly different between rheumatology fellow and medical student groups. The multivariate logistic regression model of the literature factor model indicated much poorer fit than with the literature model with component individual responses (AIC 139.555 versus 111.167) (Table 3), indicating that the a priori grouping did not improve fit. Stepwise selection found slightly worse model fit (AIC 140.445) and found selected composite measures for time off (OR 1.160, P = 0.0237) and training issues (OR 0.728, P = 0.0045) as statistically significant. Therefore, the original literature-derived factor model did not fit the data as well as the component response model.

We used factor analysis to address an empirical gathering and weighting of response data into underlying factors. Table 4 lists the factor pattern, i.e., the correlations of individual responses with the 5 underlying or latent factors representing unmeasured factors responsible for correlated responses. Items were considered correlated if r was >0.4. Factor 1 was interpreted as “time” based on correlations with time control, number of hours, call in practice, family, and time off. Factor 2 was termed “money,” with modest correlations with likelihood of litigation, job opportunities, remuneration, training length, prestige, nonfaculty role model, and clinical role model. Factor 3 was termed “external constraints” and correlated with responses to loan burden, spouse career, difficulty in obtaining training position, own age, and life experiences. Factor 4 was termed “practice content” and correlated with patient type and continuity of care, as well as (modestly) with intellectual interest. Factor 5 was termed “academics” and correlated with intellectual interest and interest in academic medicine.

The factor analysis was then used to derive factor scores that were used in multiple logistic regression comparing rheumatology fellows with medical students. The factor score model fit with the 5 exploratory factors (AIC 99.606) was greatly improved over any other, including the original literature model, the literature plus exploratory model, or the literature factor model (Tables 3 and 4). Stepwise selection found the reduced model had improved fit with AIC 97.856, and 4 of the 5 factors were retained: time (OR 2.081, P = 0.0128), external constraints (OR 0.248, P = 0.0002), practice content (OR 2.042, P < 0.0337), and academics (OR 4.30, P = 0.0002). Therefore, when rheumatology fellows were compared with the medical student group, rheumatology fellows attached significantly greater importance to the factors termed time, practice content, and academics than did medical students. Factor 2 (money) was not statistically significant and nearly neutral in separating medical students from rheumatology fellows, and factor 3 (external constraints) was scored as significantly less important in the rheumatology fellow group (Figure 2).

Other descriptive analyses focused particularly on medical students and role modeling. Medical students attached the greatest importance to intellectual interest, family, time off, and matching to skills (Table 2). The lowest ranked motivations for student choices were hating long hours, own age, prestige, nonfaculty role models, remuneration, and educational loans. There was agreement among students, residents, and fellows that subspecialty prestige carries little importance in determining specialty selection. Based on other component correlations, the effect of role modeling was subsumed in factor 2 (money), both for nonfaculty role model and clinical role model, and therefore were entwined with monetary issues. Yet the univariate rankings did not place either role-model response among the top 10 items in importance, with the single exception among nonrheumatology fellows at ninth place, nor was there a univariate significant difference about role modeling between rheumatology fellows and any other group (medical student, internal medicine resident, or nonrheumatology fellow). While loan burden was not a highly ranked response item in any group, it correlated moderately with money (loading 0.69) (Table 4).


  1. Top of page
  2. Abstract

This study showed that the rheumatology fellow group attached significantly more importance to the variables representing time, practice content, and academics, and significantly less importance to external constraints, than the medical student group. Money was identified as a factor but did not differ between the rheumatology fellow and medical student groups. The values and motivations correlated with rheumatology fellowship are therefore complex. This study confirms the role of intellectual interest (part of the underlying variable that includes patient type and continuity of care, as well as part of the academics variable) (12) as one influence distinguishing rheumatology fellows from trainees at earlier stages in their education, as both students and residents (Figure 2). Therefore, one might surmise that rheumatology fellows identify intellectual interest both in rheumatic disease practice, as well as in academic medical pursuits. Therefore, by emphasizing a “bench to bedside” message, the recruiting efforts of the ACR may correctly target the intellectual interests of current fellows.

The potential biases and limitations of the current study include: 1) defects of self-reporting, 2) the possible instability in views on money or other major factors over time and stages of training, 3) the possible failure of the questionnaire to fully include or explore a major factor such as role modeling, and 4) heterogeneity among rheumatology fellows. Also, the fellows and comparison groups were geographically dissimilar. A national, comprehensive sample of rheumatology trainees was compared with local samples of other trainees. Because the respondents contained a small sample from subspecialty fellowships other than rheumatology, the data could not be used to address questions related to rheumatology versus other internal medicine subspecialty fellowships. It is likely, for example, that all medicine subspecialty fellows, not just rheumatology fellows, manifest the characteristics described here, such as intellectual interest in their area. It is also likely that internal medicine trainees most concerned with debt are missed by studies such as this one that omit procedural subspecialties. The students and fellows are at entirely different stages in their life/training. Additional and necessary comparisons of rheumatology fellows with other specialty fellows at the same level of training remain to be made. This comparison might be especially revealing if a large base of procedurally-based fellows is included. Under current economic conditions, the relative importance of controllable lifestyle, time (Figure 2), and money-related items may shift. The same might happen with the balance between intellectual interest and money. However, we believe, based upon the literature that in 1989 first identified the increasing influence of controllable lifestyle (10), that these balances are determined by powerful underlying demographic and social factors that are irreversible and that will outlast even large fluctuations in economic conditions. Examples of potent factors include the evolving equality of sex roles and 2-income families among physicians.

One might ask what practical lessons derive from our findings for rheumatology fellowship programs. Despite the limitations, the factor analysis suggests that the considerations that go into choosing rheumatology or other career paths can be organized around major variables representing time, money, practice content, external constraints, and academics. Rheumatology already has a strong position within the time and practice content domains; trainees commonly see that time allocation and practice are significant advantages. One intervention to enhance rheumatology fellowship might be to foster early exposure and education about specific important underlying factors. A clinician can point out, and foster by clinical experience, the positive aspects of rheumatology with regard to patient profile, controllable lifestyle, and intellectual rewards. This might mean, for example, mentored rheumatology electives during internship. Also, because the magnitude of effects is quite substantial for the external constraints factor, a second direction for intervention would be to mitigate influence of underlying factors that may tend to prevent rheumatology fellowship (i.e., those factors with ORs <1.0 in this study). One would, therefore, improve external constraints (factor 3) by minimizing, insofar as possible, the negative effects from loan burden, spouse career, and difficulty in obtaining a fellowship position. Apart from loan burden, the external constraints domain contains few other modifiable elements; one cannot intervene meaningfully regarding spouse career, age, or life experience. Therefore, targeted improvement within the external constraints factor (Figure 2) would have to be financial, through loan burden and availability of training position. The timing would need to be early enough, such as within the first year or two of postgraduate education, to affect career decisions.

Based on these findings, we hypothesize that the most effective recruiting interventions would deliver more practice content and useful information to trainees before the late internship recruiting window. For example, in rheumatology, information delivery before the internship recruiting window could be by earlier exposure through curriculum, as suggested by Kolasinski et al (12). The failure of this statistical analysis to emphasize the importance of role modeling does not mean it was unimportant, but rather equal in importance to both groups: medical students and rheumatology fellows. Timing of recruiting may be just as important as picking the right factors to emphasize. Kolasinski et al also found that 75% of 177 rheumatology fellows chose their field during late internship or early residency. Those authors also assert that, short of increasing salaries or lessening loan burden, emphasizing role modeling and curriculum efforts aimed at the internship and residency decision-making window (PGY 1 and early PGY 2) may be the most important tools available to increase rheumatology recruiting (12).

The same analysis might have far different, and helpful, implications for other US medical work force problems (5). For example, the primary care physician shortage has continued despite loan forgiveness programs; the proportion of US allopathic medical school graduates planning careers in primary care decreased from 53.4% in 1997 to 35.1% in 2004 (14). The model presented here (Figure 2) suggests that recruiting could be improved by targeting negatively-perceived domains. Practice content or time interventions such as restructuring on-call work, paperwork reduction, and increased autonomy could attract trainees more than loan forgiveness or other monetary incentives alone.

In conclusion, the analysis yielded 4 main significant factors that influence trainees toward a career choice: time, practice content, external constraints, and academics. Intellectual interest is a key rheumatology item that correlates with both practice content and academics. Money was the next (5th) most important factor, but it failed to reach significance in this analysis (Figure 2). The clustering of important items within the factors suggests the hypothesis that recruiting efforts will have the best chance for success when aimed at making high-value improvements in career items within the domain perceived as the least attractive by physicians in training. Reduced to an aphorism of recruiting, whether for general medicine or for rheumatology, conclusions derived from these data might be formulated as, “Find out what people value most, and honor it where you can.” Clearly, rheumatology fellows seem to value controllable lifestyle, practice content, and academics more than money, and external constraints issues may prevent or reduce inclination toward rheumatology fellowship.


  1. Top of page
  2. Abstract

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Roberts had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Moxley, Waterhouse, Roberts.

Acquisition of data. Carleton, Barrett, Brannen, Roberts.

Analysis and interpretation of data. Rahbar, Moxley, Carleton, Thacker, Waterhouse, Roberts.


  1. Top of page
  2. Abstract