Health Workforce Remuneration Comparing Wage Levels, Ranking and Dispersion of 16 Occupational Groups in 20 Countries

This article focuses on remuneration in the Human Resources for Health (HRH), comparing wage levels, ranking and dispersion of 16 HRH occupations in 20 countries (Argentina, Belarus, Belgium, Brazil, Chile, Colombia, Czech Republic, Finland, Germany, India, Mexico, Netherlands, Poland, Russian Federation, South-Africa, Spain, Sweden, Ukraine, United Kingdom, United States). Research questions asked are to what extent are the wage rankings, wage dispersion, and standardized wage levels are similar between the 16 occupational groups in the HRH workforce across countries. The pooled data from the continuous, worldwide, multilingual WageIndicator web-survey between 2008 and 2011Q1 have been analysed (N=38,799). Hourly wages expressed in standardized USD, all controlled for PPP and then indexed to 2011 levels. The fi ndings show that the Medical Doctors have overall the highest median wages and they have so in 11 of 20 countries, while the Personal Care Workers have overall lowest wages and they have so in 9 of 20 countries. Health Care Managers lower earnings than Medical Doctors, but in 5 of 20 countries they have higher earnings (BLR, CZE, POL, RUS, UKR). The wage levels of Nursing & Midwifery Professionals vary largely across countries. The correlation of the overall ranking to the national ranking is more than .7 in 7 of 20 countries. The wage dispersion is defi ned as the ratio of the highest to the lowest median earnings in an occupation in a country. It is highest in Brazil (7.0), and lowest in Sweden, Germany, Poland, and Argentina. When comparing wage levels in occupations across countries, the largest wage differences for the Medical Doctors: the Ukraine doctor earns 19 times less compared to the US doctor. A correlation between country-level earnings and wage differentials across countries reveals that the higher the median wages in an occupation, the higher the wage difference across countries (r=.9). In conclusion, this article breaks new ground by investigating for the first time the wage levels, ranking and dispersion of occupational groups in the HRH workforce across countries. Findings illustrate that the assumption of similarity in cross-country wage ranking, wage dispersion, and purchasing power adjusted wage levels does not hold. These findings help to explain the complexity of migratory paths seen.


Background
Wages are commonly perceived as a key factor affecting job satisfaction, retention, and attrition or migration of health care professionals within and across countries (Ferrinho et al, 1998;Dovlo, 2002;Smigelskas and Padaiga, 2007;Nguyen et al, 2008). A major problem preventing progress on insight into the relative importance of wage information in health workforce strengthening is the lack of detailed information about the wide range of health workers' occupations (De Vries and Tijdens, 2010). Typically, international databases employ high levels of occupational aggregation and are insuffi ciently standardized in their classifi cations to allow for cross-country comparability (Dräger et al, 2006). For example, while the October Inquiry and the Occupational Wages (OWW) database of the International Labour Organisation (ILO) is an important resource, for the health sector only seven occupations are included: general physician, dentist, professional nurse, auxiliary nurse, physiotherapist, medical x-ray technician and ambulance driver. Another major source, the Luxembourg Income and Employment Study, has surveyed 30 countries over the past decades, yet lacks suffi cient specifi city as most labour force surveys do not provide further detail than a 2-digit coding of ILO's International Standard Classifi cation of Occupations (ISCO). An investigation for a number of European countries concludes that no cross-country comparable data is available for the occupational groups in the HRH workforce, and that one has to rely on a few national studies with incomparable wage data and incomparable occupational groups (Pillinger, 2010). At the country level, a small diversity of HRH sources is available and includes population censuses and surveys, facility assessments, and routine administrative records. However, most available data sources have shortcomings (Dal Poz et al, 2009;Mc-Quide et al, 2009).
As a result of this absence of comparable wage data, few studies have investigated wage levels and wage distribution across countries (Dräger et al, 2006;Vujicic, 2004). Preliminary analysis has suggested that salary differentials between source and destination countries are too high to curb migration (Vujicic, 2004). Using data on 42 countries from both the OECD Health Data 2005 and OWW database for a comparison of wages of general physicians and professional nurses only, Dräger et al. found that there is an enormous gap in wages for health workers between rich and poor countries (Dräger et al, 2006). Moreover, health workers tend to be paid less than equivalent professionals -or at least teachers and engineers -in low-income countries. Wages, they suggest, are great incentives for health workers to migrate, posing challenges for the development of strategies to retain them in poor countries. At the same time, an increasingly complex remuneration landscape in destination countries is showing the development of different task profi les and related certifi cations requirements-a proxy for relative wage ranking for distinct occupations across countries-across counties (Grimshaw and Carroll, 2008;Jaehrling, 2008) This article introduces a non-probability dataset that can be used for comparing wage information across countries -the WageIndicator web-survey-with the aim of contributing to an improved understanding of global wage differentials, thereby illustrating the usefulness of online data collection for crosscountry comparative research. The paper focuses on the validity of three wage cross-country wage assumptions: the similarity of ranking of wage levels, the similarity of wage dispersion and the comparability of cost-of-living adjusted (PPP) wage levels across 20 countries in 16 occupational groups in the Human Resources for Health (HRH) workforce. Using detailed occupational wage information available from the international, multilingual WageIndicator web-survey in these countries, the following three research questions will be answered: 1) To what extent are the rankings between the 16 occupational groups in the HRH workforce similar across countries, based on their median wage levels?
2) To what extent are countries similar with respect to the wage dispersion across the national HRH occupations?
3) To what extent are the standardized wage levels within the same HRH occupations comparable across countries?
Answers to these questions are important. Differences between the complexity of wage structures between various countries are of potential key infl uence to workforce migratory patterns, while allowing insight in possible strategies to increase country level job satisfaction and retention, and national settings of health care provision, wage setting processes, and credentialism.

Data
The data used in this paper stem from the WageIndicator web-survey (www.wageindicator.org). This is a multi-country, continuous surveys, posted at the WageIndicator websites in an increasing number of countries.
In 2000, the WageIndicator project started as a paper-and pencil survey for establishing a website with salary information for women's occupations in the Netherlands, but quickly developed into an online, multilingual data collection tool which on an ongoing basis pulls occupational information for hundreds of occupations through more than 60 national websites as of early 2011. A national website hosting the survey tool consists of job related content, labour law and minimum wage information, an anonymous questionnaire with a prize incentive, and a free and crowd-pulling Salary Check presenting average wages for occupations based on data from the questionnaire. Additionally, the project includes search engine optimization, webmarketing, publicity, and answering visitors' email. Most countries have their own web-manager. Coalitions with media groups and publishing houses with a strong Internet presence contribute to the large numbers of visitors to the websites. The websites are consulted by employees, students, job seekers, individuals with a job on the side, and alike for their job mobility decisions, annual performance talks, occupational choice or other reasons. All web-visitors are asked to complete voluntarily the web-survey, in return to the free information provided. Importantly, approximately 1.5% of the visitors start completing the questionnaire.
The web-survey is comparable across countries, it is in the national language(s) and it has questions about wages, education, occupation, industry, socio-demographics, and alike . The survey has a prize incentive and it takes approximately 10 minutes to complete part 1 and 10 minutes for part 2.
From a scientifi c perspective, concerns have been raised in relation to the quality and reliability of web-survey data (Couper, 2000). The problem of sample bias arises when those not covered, not recruited, and/or not surveyed are different from those who are covered, are recruited and have responded (Groves, 2004). To minimize such bias, researchers have traditionally attempted to create samples that provide a reliable cross-section of a given population allowing the drawing of probability-based samples which produce representative results for the entire population. In the case of the WageIndicator web-survey, which is a non-probability or volunteer survey, the most serious problem is related to the self-selection recruitment method of respondents, and the related question of to what extent the results are representative for the general population. To deal with this problem, different weighting techniques have been proposed to adjust a "biased" web sample to the population under consideration (Lee and Vaillant, 2009;Schonlau et al, 2009).
The effi ciency of different weights in adjusting biases has also been considered in the case of the WageIndicator data (Steinmetz and Tijdens, 2009). Specifi cally, un-weighted and weighted results of these data from the year 2006 for selected countries (Germany, the Netherlands, Spain, the US, Argentina and Brazil) have been compared using representative reference surveys for the same year. Similar to fi ndings from previous studies (Lee, 2006;Loosveldt and Sonck, 2006), the results showed that all web samples deviated from the reference samples with regard to the common variables age, gender and education. However, the impact of the applied weights seems to be very limited and does not make web-survey data more comparable to the general population. This argument can also be supported by a detailed comparison of the WageIndicator data to other so-called representative surveys (such as the Labor Force Survey or the World Values Survey) using the distributions over 36 categories (2genders*2workinghours*3agegroups*3educationgroups). As shown in their analysis (Steinmetz and Tijdens, 2009), for most of these categories it would be exaggerated to speak of a fundamental selection bias in the case of the volunteer data set. It seems worthwhile to emphasize the argument made by Couper and Miller (2008) that it is better not to treat survey quality as an absolute, but to evaluate quality relative to other features of the research design and the stated goals of the survey.

Defi ning health sector occupations
The WageIndicatorweb-survey asks in detail about the occupation of the respondent, offering a search tree with some 1,700 occupations, coded according to ILO's recently updated occupational classifi cation ISCO-08, adding further digits to its 433 four-digit occupational units . These 1,700 oc-  Table 1, given the selection of countries discussed in the next section.

Defi ning wages
The WageIndicator web-survey asks respondents about their earnings . In the survey, the employees and the self-employed are routed differently through the pages with questions on wages. The employees are asked if they are paid per month or per week, whichever is most common in the country of survey. If the answer is 'no', the next question asks them to tick the pay period. In countries where it is deemed necessary, a question asks about the currency in which the wage is paid. Then, the employees are asked 'Do you know your gross and your net wage?'. Depending on the answer, questions follow for the last gross and/or net wage. Here, a hint suggests to include bonuses, if these were received in the last wage. The next page presents a list of bonuses and benefi ts that may have been included in the last wage, ranging from shift and commuting allowances to tips and performance bonuses. These questions are default set to 'no'. If 'yes' is selected, a question pops up asking for the amount of the bonus. The self-employed receive a question about their gross annual income, followed by a question whether this income was earned in 12 months or less, and if less, in how many months. For the computation of the hourly wages, either the contractual hours for workers in dependent employment with agreed working hours in their employment contract are used or the usual working hours for all other categories. The wage variable is taken from the survey question about gross wage or net wage, which have been tested against the minimum and maximum values, applicable for the country and for the reported pay period. Then the total of reported bonuses is deducted from the reported wages. Next, the hourly wages are computed from the weekly hours, the wage period and the gross wages minus the bonuses. For the cases with information about net hourly wages only, the gross hourly wages are computed based on the annual country average between gross and net wages.
We then converted the hourly wages into a standardized hourly wage in US dollars, using purchasing power parities (PPP) from the World Bank Database with their projections for the years up to 2011. The purchasing power parity theory uses the long-term equilibrium exchange rate of two currencies to equalize their purchasing power for a given basket of goods. Using a PPP basis is arguably more useful when comparing differences in living standards on the whole between nations because PPP takes into account the relative cost of living and the infl ation rates of different countries, rather than just a nominal Gross Domestic Product (GDP) comparison. In the data cleaning, the standardized hourly wages are tested for their reliability. Indexed hourly wages lower than 1 standardized PPP US dollar or over 400 standardized PPP US dollars are considered outliers. Odd values in the reported gross and/or net wages are set to missing.
Similarly, this is done if the sum of bonuses is larger than 2/3 of the reported gross wage, or if the reported gross wages are larger than 100 times the reported net wage. In case an HRH occupation in a country had less than 5 observations over these years, the wages in this occupation were set to missing. In the remaining, the words standardized USD wages will be used to refer to the PPP standardized wages in US dollars, indexed to the 2011 level.

Wage rankings of occupations across countries
The fi rst research objective addressed to what extent the wage rankings for the 16 occupational groups in the HRH workforce are similar across countries. For this purpose, the median wages of the 16 occupations in each of the 20 countries have been computed and ranked. Ranking runs from 1, indicating the occupation with the lowest median wage in the country, to 16, indicating the occupation with the highest median wage in the country. In a few countries, wage information for some occupations had insuffi cient observations (<5), for example for the Dentists (insuffi cient in 7 countries), the Physiotherapists (in 6 countries), and the Personal Care Workers in Health Services (in 6 countries). In these countries, the ranking of these less than 16 occupations was scaled between 1 and 16. The ranking of 16 occupations in each of the 20 countries can be found in Appendix 1. Based on the median standardized wages of each occupation in each country, the 20-country mean standardized wages were calculated and subsequently ranked (Table   3, column 2 and 3). Note that this ranking does neither control for the relative sizes of the national HRH workforces nor for the relative sizes of the HRH occupations within the country. Thus, the ranking is based on occupations, not on jobholders in occupations. The results are shown in Table 3.   3 shows, not surprisingly, that the occupational group Medical Doctors rank the highest number 16, indicating that this occupational group has the highest mean across the 20 countries of the countryspecifi c median standardized USD wages. It has the highest median wage in 11 of the 20 countries and the one-highest in another three countries (see Appendix 1). The Medical Doctors group ranks relatively low in the Ukraine. The Dentists group is ranked 15 across the 20 countries, but this occupation has the highest median wage in three countries (Belgium, Netherlands, United Kingdom).
In contrast to the Medical Doctors group, the Personal Care Workers group is ranked 1, indicating that in the 20 countries this group has the lowest wage ranking, when averaging the median wages in this occupation across the 20 countries. In 9 of the 20 countries this occupation indeed ranks lowest, and in the other countries it is ranked among the lowest earning occupations, apart from the Czech Republic (rank 15), Colombia and Ukraine (both rank 10).
In most countries, the Health Care Managers group has a relative high ranking, though in three countries this occupation ranks in the middle, namely in Spain, Germany, and India. In almost all countries, the Health Professionals. Yet, in two countries the latter occupation has higher median earnings than the former, namely in South Africa and United Kingdom. The distinction between the two occupational groups is probably not understood the same way in these countries. This certainly calls for further investigations of the work activities associated with these occupational groups.
Research objective 1 aimed to investigate to what extent the rankings of the median wage levels of the 16 occupational groups in the HRH workforce are similar across countries. For this purpose, the ranking in each country has been correlated to the overall 20-country ranking, thereby indicating how much the country's ranking fi ts into the overall ranking. The one-last column in Table 4 shows the results. It depicts that the correlations are pretty high for most countries. In seven countries (Argentina, Belgium, Brazil, Chile, Finland, Netherlands, United States) the correlation is more than .7. In another four countries it is between .5 and .7 (Germany, Spain, South Africa, United Kingdom). In seven countries it is between .3 and .5 (Belarus, Colombia, Czech Republic, Mexico, Poland, Russian Federation, and Sweden). Finally, two countries exhibit a ranking that is extremely different from the overall 20-countryranking, namely India, and Ukraine.
In conclusion, for the majority of countries in this study, the ranking is pretty similar. These countries are seemingly a group of higher income countries, somewhat contrasting with the medium or lower level income countries showing a lower ranking of median wage level. Considering health workforce migratory patterns from low to higher level countries, this is difference may be of further interest.

Wage dispersion within countries
Research objective 2 aimed to investigate to what extent countries differ with respect to the gap between the highest and the lowest earning occupation in the national HRH workforce. The results are shown in Table 4. Per country, columns 2 and 3 reveal the lowest and highest median standardized hourly wages of the 16 occupations. Column 4 shows the ratio between the highest and lowest wages. This column reveals that the wage gap is largest in Brazil where the median wage of the highest paid HRH occupation is 7.0 times the median of the lowest paid HRH occupation, followed by Czech Republic, United States and Russian Federation (ratios between 5.0 and 6.3). In contrast, Sweden, Germany, Poland, and Argentina are egalitarian countries as far as the median wages in the HRH workforce is concerned (ratios between 1.6 and 2.5).
In another fi ve countries, the ratios are between 3.0 and 3.5 (Spain, Belarus, Ukraine, United Kingdom, and Mexico). In the remaining six countries, the wage differentials are between 4.0 and 4.9 (Netherlands, South Africa, Colombia, Belgium, Chile, and India). One can conclude tentatively that wage dispersion is higher in the larger economies, such as Brazil, United States and Russia, compared to smaller economies, but that in general a diverse pattern is seen.    Table 3 shows that these are lowest in the Russian Federation, Ukraine, Czech Republic, Brazil, India, and Poland. They are highest in Germany, Sweden, the United Kingdom, and the Netherlands. Across countries, the wage differentials within occupations are highest for the group of Medical Doctors and lowest for the group of Personal Care Workers (Graph 2). Across countries, the mean wages within-occupations -thus the sum of the median wages in this occupational group divided by the number of countries with valid wage data for this group -are highest for the group of Medical Doctors and lowest for the group of Personal Care Workers (Graph 2). The correlation between the within-occupation wage differentials and the within-occupation mean wages is high (r=.9), indicating that the wage distributions in the health workforce reveal similar patterns across countries. Assuming that workforce mobility across countries is driven by wage differentials, provided that these wage differentials are perceived to be controlled for PPP, one can expect the groups of Medical Doctors and Dentists to migrate from the former Eastern European countries to the UK, US, South-Africa and the Netherlands. Based on the wage differentials in Graph 2, lower workforce mobility though still substantial can be expected for the remaining occupational groups. The lowest mobility can be expected for the occupational group of the Personal Care Workers.

Discussion
This study certainly has limitations. The fi rst one relates to the defi nition of wages. WageIndicator applies a standard defi nition to all countries and occupations, as explained in section 3. However, wage structures may vary across countries. It may include non-fi nancial remunerations such as housing or food, may include fi nancial remunerations probably not reported as wage such as transportation cost reimbursement, may include social benefi t or pension contributions, or may include in part cash rewards not reported. Thus, whereas the web-survey has a standardized approach of calculating hourly wages, there may be variation across countries which are not taken into account. Possibly this would explain the fi nding that median wages for Associate Nurses and Midwives wages are higher than Nurses and Midwives in the United Kingdom and South Africa.
A second limitation relates to the occupational titles. In this study, it is assumed that the same occupational titles to refer to the same job content across countries. Thus, the occupational group of Nursing & Midwifery Professionals is assumed to have the same set of tasks across the world, otherwise the wages of apples and pears would be compared. However, the job content of the HRH occupational groups is not empirically tested on a worldwide scale. The WageIndicator web-survey does allow for a worldwide testing of job content, but this would require a separate project for developing such testing.
A third limitation relates to the diploma credentials in the HRH occupations. In most countries for most HRH occupations credentials are required. Depending on the supply and demand ratio in the local labor market, these credentials will or will not be required for entry into the job. In most workplaces credentials will lead to higher earnings. However, the current dataset does not allow controlling for credentials. Thus, the dataset does not control for wages of accredited versus not-accredited jobholders in the same occupational group.
Finally, this study does not take into account the public or private provision of health care, which is assumed to affect wage setting. It also does not take into account regional wage differentials in large countries.
Nevertheless these limitations, being the fi rst study on wages in a wide range of HRH occupations and a wide range of countries in four continents, it certainly increases the understanding of wage levels and wage dispersion in the HRH fi eld.

Conclusions
This paper breaks new ground by investigating for the fi rst time the wage levels and the wage distribution of 16 occupational groups in the Human Resources for Health (HRH) workforce for 20 countries.
Cross-country worldwide wage comparisons have not been undertaken for such a great detail in occupational breakdown. This data is needed for understanding cross-country mobility in the HRH workforce, for understanding the national settings of health care provision, and for understanding wage setting processes and credentialism within countries. For the analyses, the wages were fi rst controlled for purchasing power parity in the respective years, and then these wages were set to the 2011 level. In total, the analyses included38,799 observations.
Research question 1 assumed that the ranking of median wages in the 16 occupational groups was similar across the 20 countries. The study reveals that in the majority of the countries the wage ranking is indeed fairly similar across countries, particularly for higher income countries. In 7 of the 20 countries, the national ranking correlates at least .7 with the overall 20-country ranking. The fi ndings show that the Medical Doc- Research question 2 assumed that the wage distribution among the 16 occupations was similar cross countries. This assumption did not hold. The wage dispersion is defi ned as the ratio of the highest to the lowest median earnings in an occupation in a country. It is highest in Brazil (7.0), whereas Sweden, Germany, Poland, and Argentina are egalitarian countries as far as the median wages in the HRH workforce is concerned.
Research question 3 assumed that the wage levels within the same occupational groups in the HRH workforce were comparable across countries, using standardized PPP wages. The largest wage differences are found for the Medical Doctors: the Ukraine doctor earns 19 times less compared to the US doctor.
Correlation between country-level earnings and wage differentials across countries, the data reveal that the higher the median wages in an occupation, the higher the wage difference across countries (r=.9).