| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||
RESEARCH ARTICLE |
1 Interdisciplinary Research Centre on Ageing, Swansea University, United Kingdom.
2 Institute of Medical and Social Care Research, University of Wales, Bangor, United Kingdom.
3 Faculté des Lettres, des Sciences Humaines, des Arts et des Sciences de l'éducation, Université du Luxembourg, Luxembourg.
4 Department of Cognitive Science and Education, University of Trento, Italy.
5 Blekinge Institute of Technology, School of Health Science, Karlskrona, Sweden.
6 Amsterdam Institute for Metropolitan and International Development Studies, Universiteit van Amsterdam, The Netherlands.
7 Faculty of Psychology, Department of Clinical, Biological and Differential Psychology, University of Vienna, Austria.
Address correspondence to Vanessa Burholt, Interdisciplinary Research Centre on Ageing, School of Human Science, Swansea University, Singleton Park, Swansea, Wales, United Kingdom SA2 8PP. E-mail: v.burholt{at}swansea.ac.uk
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionAppendixReferences |
|---|
Methods. A questionnaire was administered through face-to-face interviews in five countries, and postal interview in the sixth, to representative populations of adults aged 50 to 90 living independently (N = 12,478). This article examines missing values and distribution of items in the social resources scale, and consistency of skew and kurtosis across countries. We performed item–total correlations and ran confirmatory factor analyses to test a three-factor model obtained in previous U.S. and Spanish analyses. Cronbach's alpha determined the reliability of the factors.
Results. We observed a relatively large proportion of missing data for one item (have someone who would help you). All items correlated with a score equal to or greater than 0.20. Although the confirmatory factor analyses generally supported the acceptability of the three-factor structure in the European data, the reliability of two dimensions (dependability and affective) was unacceptably low.
Discussion. Differences across countries make it unlikely that researchers can develop a single social resources scale that would have item equivalence in multiple countries.
DESPITE the considerable accumulation of research on social relationships, there is little agreement on the best methods of measuring and conceptualizing social resources. This is partly due to the vast array of measures that have been used (Glass, Mendes de Leon, Seeman, & Berkman, 1997
; Hanson & Östergren, 1987; Heitzmann & Kaplan, 1988).
Research has demonstrated that social resources have multiple dimensions (Hall & Wellman, 1985; Wellman, 1988) and for older people are the products of the cumulative impact of life course factors (Connidis & Davies, 1990). Measurements of social resources often distinguish between the structure and the function of the relationships within the network of social resources (Barrera, 1986; Hanson & Östergren, 1987; House & Kahn, 1985). Structure includes the frequency and quantity of social contact with other individuals (Avlund, Damscaard, & Holstein, 1998; Cooper, Arber, Fee, & Ginn, 1999), whereas function is often described in terms of the type of support provided (e.g., emotional, informational, or practical; Avlund et al., 1998; Cohen & Willis, 1985; House & Kahn, 1985; Kessler, Price, & Worman, 1985). Some have argued that the quality rather than quantity of social resources is likely to exert an impact on an individual (Orth-Gomér & Undén, 1987; Pinquart & Sörensen, 2000). Therefore, measurement of social resources should also take into account satisfaction expressed about the quality and quantity of relationships (Cooper et al., 1999).
Criticism of social support research has focused on the lack of a single measure that captures the dimensions of social support and can quantify the intensity of support (Madge & Marmot, 1987). Other criticisms of social support network measurement techniques have focused on the lack of reporting on the psychometric properties of scales and the fact that there is little or no reporting of validity and reliability (O'Reilly, 1988; Winemiller, Mitchell, Sutliff, & Cline, 1993). A systematic review of 17 existing models for social resources showed that although functional measures tended to be multidimensional, they were difficult to administer, time consuming, and difficult for respondents to understand. In contrast, structural measures were less time consuming and were easily understood by respondents but were not assessed for validity and reliability. Overall, scholars have suggested that the "ideal study" include structural and functional measures of social support (Orth-Gomér & Undén, 1987).
In the European Study of Adult Well-Being (ESAW), a selection of questions addressing the structure, function, and quality of contact with others was used to rate social support resources according to the Older Americans Resources and Services (OARS) Multidimensional Functional Assessment Questionnaire (see Appendix). The Duke University Center for the Study of Aging and Human Development developed OARS, which includes an assessment of personal functioning in five domains: social, economic, mental health, physical health, and self-care capacity. A summary rating is calculated for each of the domains that ranges from excellent functioning (1) to totally impaired (6) (Fillenbaum & Smyer, 1981). Previously, a test–retest reliability trial showed that 91% of items were identical after a 5-week interval, and an intrarater reliability trial demonstrated that 80% of intrarater correlations were 0.8 or higher (Fillenbaum, 1988). Although other domains in OARS have been previously assessed for validity, the social resource component was not examined, as an external standard of comparison was not identified (Fillenbaum & Smyer, 1981).
In the development of the OARS social resources scale, researchers undertook factor analysis of the items in the social resource domain with a sample of older Americans (N = 2,036) (Fillenbaum, 1988). They identified three factors that reflected the availability and amount of contact with friends (the interaction dimension), the availability of close support (the dependability dimension), and the adequacy of contacts (the affective dimension). A Spanish study replicated the factor groupings with a sample of 473 older people (aged 60 and older) living in Granada, Malaga, Cordoba, and Seville (Fibla, Patiño, & Domínguez, 1996). We discuss the reported internal consistency (reliability) of the dimensions identified by the factor analysis later in the article.
The purpose of this article is to examine data quality, reliability, and construct validity of the OARS social resources scale in six European countries. We used confirmatory factor analysis to test whether the three-factor solution identified in the United States and Spain was consistent with the European data. The three dimensions of the OARS social resource scale are not used as individual subscales in analysis. Therefore, we also explored the data quality of the global social resources scale.
|
METHODS |
|---|
|
TOPAbstractMethodsResultsDiscussionAppendixReferences |
|---|
). The United Kingdom initiated and coordinated the ESAW research. A principal coordinator and team of researchers conducted the work in each participating country.
Research Design
Questionnaire development
Respondents in the participating countries used nine languages. Researchers at Indiana University had prepared a questionnaire for the global project in American English. The ESAW participants produced compatible versions of the questionnaire for each country and translated the questionnaire into the necessary languages using the translation/back-translation procedure to ensure semantic equivalence (Bowden & Fox-Rushby, 2003). It was important that universal meaning was shared across the cultural groups participating. As recommended by Herdman, Fox-Rushby, and Badia (1998), prior to translation a review of the literature and consultation with the multidisciplinary project teams (including an anthropologist, sociologists, psychologists, health care professionals, and geographers) from all six countries established that there was evidence of conceptual equivalence. That is, the underlying concepts (domains) were applicable and acceptable to participating countries.
With regard to the social support domain, the prevailing literature suggested that there would be differences in item equivalence between countries (Herdman et al., 1998). For example there were likely to be differences in the importance of family relationships between countries, as evidence suggests that although coresidence is a welcome option in several southern European countries (e.g., Italy), it is less acceptable in Scandinavian countries (e.g., Sweden; de Jong Gierveld, de Valk, & Blommesteijn, 2001). However, the researchers decided that the measures would be relevant in all countries, albeit to a greater or lesser extent. Therefore, researchers retained all items of the OARS social resource scale without modification (other than translation).
Sample selection
Each of the country teams selected a representative population of adults aged 50 to 90 living independently (not in hospitals, residential care, or nursing homes). The sample included the indigenous population and migrants. Samples were drawn from both rural and urban areas. Because of differences in settlement patterns and population density, each participating country determined the definition of rurality according to the relevant definition applied at a national level.
In the United Kingdom, three distinct subsamples were drawn representing England, Scotland, and Wales, with one rural and one urban community chosen from each country. The Austrian sample was drawn from urban and rural districts located in each of the nine federal states constituting the country. In The Netherlands, respondents were drawn from three rural and three urban communities in each of the three main regions (Central Region, the Intermediate Region, and the Peripheral Region). In Italy, the sample was drawn from one metropolitan area, one or more urban areas, and one or more rural areas in each of the four macroregions (Northwest, Northeast, Center, and South and Islands). The Swedish sample was drawn from the County of Blekinge (South Sweden). Municipalities were selected to represent a small rural community, a small city, a medium-sized city, one university city, and one large city. With the exception of the latter, the city samples included the surrounding rural communities. The Luxembourg sample was recruited from the 13 districts of the country, grouped into four subregions, of which two were rural (North and East) and two were urban (South and West).
With the exception of the United Kingdom, directories of residents were available from which the sample frames were obtained. In The Netherlands the list had to be purchased from the relevant municipalities. In the United Kingdom, the project team conducted a door-to-door census in the study areas to identify eligible interviewees. Both of these approaches were time consuming.
Initially the target sample was 2,500 for each country. However, the delays encountered in The Netherlands and the United Kingdom meant that the project fieldwork phase would have had to have been extended for these countries to meet the target. Consequently, in November 2002, the target sample was reduced to 2,000. Most countries had reached this target by the end of December 2002 (and some had exceeded this). The United Kingdom and The Netherlands continued to add to the sample until January 15, 2003, when interviewing ceased. Final response rates varied greatly, ranging from 41% to 88%. Actual samples reached values very close to the target, ranging from 1,854 to 2,417. The total European sample comprised 12,478 respondents.
All six national ESAW teams generated a proportionate stratified probability sample plan (Galtung, 1967; Hays, 1969) of the national population aged between 50 and 90 controlling for four age groups (50–59, 60–69, 70–79, 80–90) and gender (combining to form eight groups). We computed poststratification weights (e.g., Henry, 1990, p. 129; Lessler & Kalsbeek, 1992, p. 193; Sapsford, 1999, p. 32) for each of the eight age/gender groups within each of the six countries (not changing the number of cases) and used the weight function in SPSS to strengthen the representativeness of national estimates for social resources (Nie, Hadlai Hull, Jenkins, Steinbrenner, & Bent, 1975). The formula for these weights was w = pp/ps, where pp is the population proportion and ps is the sample proportion.
Interviewing
In five of the six countries, trained interviewers administered the questionnaire to respondents face to face, in the respondent's own home and in his or her first language. Although it was anticipated that operational equivalence (Herdman et al., 1998) would be achieved through all countries adopting a face-to-face interview technique, in Sweden this was not possible. During piloting of the questionnaire it became apparent that the response rates to personal interviews were very low, with respondents stating that they would prefer to complete the questionnaire alone. Therefore, Sweden conducted a postal survey. In Sweden a cover letter was sent with the questionnaire stating that respondents could obtain help in completing the questionnaire if required. In these instances, and in the case of high levels of missing data in returned questionnaires, project teams phoned respondents and offered phone or face-to-face interviews.
Data sets
Each team created a national database in SPSS. Researchers took care to ensure compatible data sets by using a standard template for all countries. Teams agreed on methods for handling data, cleaning, and performing reliability testing. The U.K. principal investigator subsequently combined cleaned data sets into an integrated data set.
Methods of Analysis
In order to establish whether the OARS social resources scale had measurement equivalence in the six participating countries, we undertook the following evaluations that have been used to assess other measures (e.g., Fibla et al., 1996; McHorney, Ware, & Rachel, 1994; Sullivan, Karlsson, & Ware, 1995): (a) tests for completeness of data, both at an item and overall scale level for the total sample and between countries; and (b) tests of rating scaling assumptions, including skew, distribution of item and scale responses, and item–total correlations corrected for overlap.
In addition, we also conducted (a) confirmatory factor analyses to test the goodness of fit of the hypothesized factor structures (see Figure 1) in the U.S./Spanish measurement model (Fibla et al., 1996; Fillenbaum, 1988) to the total European sample, and (b) estimation of internal consistency reliability for the underlying dimensions of the scale.
|
). We coded item raw scores and transformed them into a scale using the appropriate SPSS syntax, with ranges from excellent functioning (1) to totally impaired (6)(Fillenbaum, 1988). However, a score cannot be confidently estimated if there are high levels of missing data at the item level (McHorney et al., 1994). Therefore, we examined each item for the proportion of respondents with missing data for the total sample and for each country. We excluded respondents with missing data on any items from analysis of the social support scale rating, and we noted the percentage of respondents for whom scores could not be calculated.
Scaling assumptions
We examined the skew and kurtosis of items. The ratio of skewness (or kurtosis) to its standard error is used as a test of normality. Normality is rejected if the ratio is less than –2 or greater than 2.
Tests of item internal consistency looked at the extent to which each item contributes to the measure (i.e., the item–total correlation using Pearson product-moment correlation; Nunnally, 1970). Satisfactory reliability entails each item correlating with the scale above 0.20, with items not achieving these levels being discarded (Kline, 1986). We corrected correlations between the individual items and the scale total for overlap (Howard & Forehand, 1962).
Confirmatory factor analysis
We tested the factor structure of the OARS social resources measurement model using LISREL. We randomly split the sample using the algorithm available in SPSS to produce two samples: a calibration sample to develop the model, and a validation sample to test the derived model (Diamantopoulos & Sigaw, 2000). The development sample contained 6,286 and the validation sample contained 6,204 observations. Listwise deletion of cases with missing values resulted in effective sample sizes of 5,760 and 5,746, respectively. The specified matrices were asymptotic (for non-normally distributed data), containing the estimated sample variances and covariances (Diamantopoulos & Sigaw, 2000). For this large sample (>2,500) of non-normally distributed data, we used maximum likelihood estimation to ascertain the loadings of the variables onto the hypothesized factors (Hu, Bentler, & Kano, 1992). However, although estimates may be acceptable, there may be problems with the corresponding chi-square statistic and the standard errors of the estimates (Bentler & Chou, 1987). Therefore, we used the more robust Satorra–Bentler scaled chi-square (Satorra & Bentler, 1988).
We examined the factorial validity for the hypothesized factor structure. The goodness-of-fit indices used to assess the fit of the measurement models to the samples were as follows:
) and sample size (Biddle, Markland, Gilbourne, Chatzisarantis, & Sparkes, 2001). Given the large sample size of this study, it was likely that the chi-square statistic would not be a very useful indicator. ). ). ; MacCallum, Browne, & Sugawara, 1996).
Internal consistency reliability
We estimated reliability coefficients by computing Cronbach's alpha. Measures with reliability of.50 (Helmsteader, 1964) to.70 (Nunnally, 1978) or greater have been recommended for the purpose of comparing groups. However, there are difficulties associated with the interpretation of alpha, as it is dependent on the magnitude of the correlation among items and the number of items in a scale (Streiner & Norman, 2003).
|
RESULTS |
|---|
|
TOPAbstractMethodsResultsDiscussionAppendixReferences |
|---|
|
Completeness of Data
Table 2 lists the item frequency distribution for the total sample and the percentage of respondents with missing data for each item. Missing value rates were low, ranging from 0.6% ("times talk with someone per week") to 6.2% ("have someone who would help you") and averaging 2.1%. The highest rates of missing values were for the last item in the sequence of questions asked for OARS social resource scale. In all countries "have someone who would help you" had the highest levels of missing values. There were, however, differences in the proportions of missing values between the countries. The highest levels were in Luxembourg and Austria. This was particularly evident for the help question, for which the data were missing for 12.5% and 7.8% of respondents, respectively.
|
Rating Scaling Assumptions
The responses for the items in Table 3 show that the distribution of item responses was skewed, with respondents tending to give favorable responses. In the total sample, for all items tests of normality were rejected. The skew was less favorable for "times visit with someone per week" (a greater proportion of people reported spending time with others 2–6 times a week, rather than daily). An examination of the data by country showed that all countries followed a similar pattern of skewness to the total sample.
|
). In the total sample, for all item tests of normality were rejected. An examination of the data by country showed that items in all countries followed a similar shape of distribution to the total sample, with some deviations. In The Netherlands and Austria, "times visit with someone per week" had a normal distribution, whereas in the United Kingdom the distribution was leptokurtic. In Sweden, the distribution of feelings of loneliness was normal, whereas in Italy it was platykurtic. In addition, the distribution of "times talk with someone per week" (normal) and availability of help (platykurtic) were different from the shape of the distribution for the total sample. Without exception we observed the full range of each item in each country.
According to the 6-point classification for the OARS social resource rating, more than one third of the sample had excellent social resources and one third had good social resources. Overall, the mean rating showed that respondents' social resources were good to mildly impaired (Table 4). As with the individual items that contributed to the scale, the distribution of responses were skewed, with respondents tending to give favorable responses. The scale had positive kurtosis (leptokurtic) for the total sample. The distribution in four country samples was also leptokurtic, however in Sweden the shape of distribution was platykurtic and in The Netherlands it was normal. Tests of normality for skew and kurtosis were rejected for the distribution of the scale in the total sample and for all country samples (with the exception of kurtosis in The Netherlands).
|
Confirmatory Factor Analysis
The results generally supported the acceptability of the three-factor structure in the European data (see Table 5), although we found a significant chi-square. However, the chi-square statistic is a direct function of sample size (Bentler & Bonett, 1980), which was inflated by the large sample used in this research. Both the root mean square error of approximation and the standardized root-mean-square residual suggested a good fit according to the criteria specified previously. Likewise, the comparative fit index also suggested an acceptable fit. All of the parameter estimates were significant, and the majority were greater than.60 (see Table 6).
|
|
; Fillenbaum, 1988). The factor corresponding with the interaction domain of social support had an overall alpha of.56 (compared with.56 in the United States and.61 in Spain), with country alphas ranging from.44 in Austria to.60 in Sweden. As noted in other scale analyses, the alphas were likely to reflect the selection of items for conceptual breadth (rather than the maximization of internal consistency) (Ryff & Keyes, 1995). Some have argued that for broadly defined constructs, reliabilities as low as.40 are acceptable (Van de Venn & Ferry, 1980). The reliability was unacceptably low for the dependability dimension (.13) and for the affective dimension (.41) of social resources. The modest alpha coefficients likely reflected the small number of items in each dimension (Streiner & Norman, 2003).
|
|
DISCUSSION |
|---|
|
TOPAbstractMethodsResultsDiscussionAppendixReferences |
|---|
, p. 331). The Methods section dealt with item equivalence, operational equivalence, and semantic equivalence. Despite our initial assertions that the measure was conceptually equivalent in the ESAW nations (see the introductory paragraphs), here we discuss measurement equivalence and revisit conceptual equivalence.
In order to assess the measurement equivalence of the OARS social resources measure, we examined reliability (internal consistency), item–total correlations, and the construct validity of underlying dimensions. We used confirmatory factor analysis to determine the applicability of the factor structure of the OARS social resources scale. The results support the factor structure identified in previous research. Although the tested model may indicate a good fit of the data, it may not necessarily represent the best model when considering the wide range of possible untried models that we could have examined (Kline, 1994). It is quite possible that there are other unexplored facets of this construct.
The low reliability of two of the underlying dimensions of the social support measure (the dependability and affective dimensions) suggest that there may be (a) alternative items that more adequately describe the underlying dimensions, or (b) alternative factors that may more accurately capture social support.
Within the dependability factor there was a low response rate for the item "have someone who would help you." This may suggest that the conceptualization of dependability is different in the European context. The question "Is there someone who would take care of you if you were sick or disabled?" attempts to determine participants' access to long-term care.
The extent to which long-term care is provided by the welfare state or families varies across cultures. For example, in the United States 1.1% of the population 65 years of age and older has neither public nor private insurance (Mold, Fryer, & Thomas, 2004). A majority of the U.S. older population holds Medicare, which covers home services, but not to the extent, or of the type, covered in Europe: Medicare pays only for medically necessary skilled home health care and does not cover "custodial care" (i.e., nonskilled care that assists with activities of daily living) (U.S. Department of Health and Human Services, 2007). Medicaid provides a range of home services for eligible patients (usually the very poor or those without any financial resources) (International Reform Monitor, 2006). In Europe, the welfare provision of social care varies according to expectations and obligations for more or less family/state provision. In the context of ESAW, in Sweden, people with social care needs expect to receive support from the state, rather than the family. In the United Kingdom, Austria, The Netherlands, and Luxembourg, services are intended to support family care and access to these may include a significant degree of discretion, whereby family support may or may not be taken into account. In the United Kingdom there is a recognized downward obligation from parents to dependent children, whereas in the other three countries there is an upward obligation for care from adult children to their parents. In Italy the state takes a noninterventionist approach, and the extended family is seen as the first source of support when members are in need (Scharf et al., 2003). Due to the limited nature of formal provision of home services in the United States and Italy, participants in surveys are likely to perceive that the question on availability of help relates to the availability of informal carer(s). At the other extreme, in Sweden availability of long-term term care is likely to refer to state sources. For the other countries, it may be unclear to participants in the survey whether they should refer to the availability of informal or formal support. The lack of clarity may have had an impact on nonresponse to this question.
We also observed a low internal reliability for the affective dimension. This may indicate problems with the theoretical conceptualization of the dimension of affect. Currently the hypothesized factor structure suggests that loneliness and satisfaction with contacts with relatives and friends contribute to this factor. Weiss (1973) suggested that two components of loneliness can be identified: emotional loneliness and social loneliness. Emotional loneliness is the absence of a significant other (e.g., a partner or best friend) with whom a close emotional attachment is formed and thus would correspond to the item relating to the presence of a confidant. Social loneliness relates to the absence of a social network consisting of a wide or broad group of friends, neighbors, and colleagues and could relate to satisfaction with contacts with relatives and friends. It may be important to consider the former item as an alternative to self-assessment of loneliness, as there is stigma attached to disclosing these feelings (Routasalo & Pitkala, 2003; Weiss, 1973).
Overall, participants in Sweden appeared to have fewer social resources than those in other countries. Observed differences between the countries appear to be due to cultural conventions with regard to living arrangements, and family and social norms. In addition to the differences with regard to expectations for care provision (see previous discussion), in Sweden, the method of socializing is more formal than in other European countries. Generally, Swedish people do not expect friends to just pop in but make formal arrangements to visit. In addition, members of a couple that are committed to a relationship with each other may choose to live apart in separate residences. In 2001, 14% of the Swedish population aged 18 to 74 who were not married or cohabiting had living apart together relationships, known as särbo in Swedish (Levin, 2004). Levels of loneliness in Sweden in the ESAW study were also comparatively high (35% were quite often or sometimes lonely). However, the measurement of this item appears to have convergent validity with a scale used in the Kungsholmen project, in which one third (33%) of nondemented participants aged 75 years and older were often lonely (Holmén, Ericsson, & Winblad, 2000). The low levels of social resources in Sweden appear to be an artifact of Swedish society rather than a measurement issue.
Cross-cultural comparisons of social resources are becoming more common, and researchers often assume the generalizability of measures across countries without evaluation of the measurement instrument. In this instance the parent project in the United States selected the OARS instrument as a robust measure to be used in international comparison. However, the European substudy indicates that there are difficulties associated with using the American measure in six different European settings. Although the interaction dimension of social support is fairly reliable, the other two dimensions are not sufficiently reliable measures of social support in the countries included in the ESAW study. Consequently, a more robust measure than the OARS social resources scale is required to measure social resources in the European context. At the outset of the study the collaborators decided that the measures would be relevant in all countries, albeit to a greater or lesser extent. However, the results suggest that differences across countries in provision of services, expectations regarding family responsibility, and norms of social contact make it unlikely that researchers can legitimately develop a single social resources scale that would have item equivalence in multiple European countries.
|
Appendix |
|---|
|
TOPAbstractMethodsResultsDiscussionAppendixReferences |
|---|
How many people do you know well enough to visit with in their homes?
About how many times did you talk to someone—friends, relatives, or others—on the telephone in the past week (either you called them or they called you)? [IF SUBJECT HAS NO PHONE, QUESTION STILL APPLIES]
How many times during the past week did you spend some time with someone who does not live with you; that is, you went to see them or they came to visit you, or you went out to do things together?
Do you have someone you can trust and confide in?
Do you find yourself feeling lonely quite often, sometimes, or almost never?
Do you see your relatives and friends as often as you want to, or not?
Is there someone who would give you any help at all if you were sick or disabled, for example your husband/wife, a member of your family, or a friend?
[IF "YES" ASK a. AND b.]
Code: Spouse = 1, Sibling = 2, Offspring = 3, Grandchild = 4, Other Kin = 5, Friend = 6, Other = 7
These items are copyright 1975, revised 1988, Center for the Study of Aging and Human
Development, Duke University. Reproduced with permission.
|
Acknowledgments |
|---|
Vanessa Burholt was PI in the United Kingdom, wrote the article, and performed some statistical analysis. Gill Windle (United Kingdom) performed the confirmatory factor analysis and wrote some statistical components. Dieter Ferring was the PI in Luxembourg and contributed to the statistical components. Cristian Balducci (Italy) contributed to the statistical components and the discussion. Cecilia Fagerström (Sweden) edited the article. Frans Thissen was the PI in The Netherlands and wrote the section on weighting the sample. Germain Weber was the PI in Austria. G Clare Wenger (United Kingdom) coordinated the study, and edited the article.
|
Footnotes |
|---|
Received for publication December 11, 2006. Accepted for publication July 30, 2007.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionAppendixReferences |
|---|
| ||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
