This study was initiated by a traditional land-owner from an Arnhem land community, who requested that researchers investigate the links between participation in natural resource management activities and human health. Ethics approval was obtained in 2004 from Charles Darwin University (H04053) and the NT Department of Health and Community Services (04/35) which includes an Aboriginal ethics sub-committee approval process. Approval was also granted from the Indigenous governed community health board and the Indigenous governed outstation resource organisation. The study setting was a large remote Indigenous community in Arnhem Land. The township, a conglomerate of 11 language groups established in 1957, is surrounded by 32 established homelands. This community has undergone a rapid transition over 50 years, becoming largely sedentary and reliant on income support. Within the population there is wide variation in caring for country participation.
Participants and procedures
Participants volunteering for the community preventive health check program were 301 Indigenous adults (177 men, 124 women) aged 15 to 54 years (M = 30.96, Sx = 10.15), comprising 23.4% of the community population in this age range . The cohort age structure differed slightly, but not significantly, from the census profile (Pearson's Chi Square statistic: 10.04, p = .19) . Of the participants, 298 (99%) completed an interviewer-administered caring for country questionnaire. The same interviewer administered the questionnaire on each occasion. Approximately one-third (N = 102) of participants came from 16 homeland communities and the remainder from the township. We undertook purposive sampling to recruit participants with different levels of involvement in formal and customary caring for country activities. Participants were from homelands, township residences, workplaces (rangers and non-rangers) and public spaces (outside the community store and community council buildings).
Three participants did not complete the caring for country questionnaire. Ten could not have weight and height recorded on the standardised equipment due to disability (N = 1) and equipment delays in the aftermath of tropical cyclone Ingrid (N = 9). Nine failed to complete questions on physical activity and diet. Five women were in the early stages of pregnancy and these participants were excluded from the final regression modelling of BMI. As there were fewer than 5% missing data for any variable, and no missing data for most variables, we imputed missing data using Full Information Maximum Likelihood estimation. Imputed means and standard deviations were identical or near-identical to those derived from the dataset with missing values; we used the imputed values for all further analyses.
Spending time on country, the seasonal burning of annual grasses, gathering of food and medicinal resources, performing ceremonies, production of artworks and protecting sacred areas are identifiable 'caring for country' activities [18, 19, 35]. Participants reported how often they participated in these six activities over the preceding twelve months on a four point ordinal response format: 1 = "Not much (none in the last year)"; 2 = "A little bit (a few days in the last year)"; 3 = "A fair bit (a few weeks in the last year)"; 4 = "Heaps (a few months in the last year)" (Additional file 1). Two further questions investigated time spent on homelands: (i) "In the last year, where did you spend most of your time living?" (the township name, homeland or other) and (ii) "How much time have you lived in a homeland/outstation in the last five years?" (all the time, a few months each year, a few weeks each year, a few days each year or none).
Follow-up and treatment were provided as clinically indicated, including a feedback letter outlining an individually tailored strategy for good health. At the time of feedback (a minimum of two weeks later) participants completed the questionnaire a second time with the same interviewer. Sixty-six participants (22% of the cohort) repeated the questionnaire within 6 weeks (M = 30.7 days, Sx = 7.99).
A senior Indigenous member of the community with well-established community links across all language groups and knowledge of all the participants also completed the questionnaire for each respondent. This 'proxy respondent' had not been involved in the health check program and had no knowledge of participant health outcomes or responses to the questionnaire. We compared the proxy's response with those of respondents', an acceptable and validated method to verify health related behaviours in remote Indigenous settings .
Caring for country questionnaire development
The questionnaire was developed in four stages over a two-year period of collaboration with an Indigenous community in Arnhem Land.
Stage 1: Scoping Study, Literature review, consultation and participant observation
Several databases covering a range of disciplines were searched for material on Indigenous caring for country and health. These included: APAIS, MEDLINE, PubMed, CINAHL, ATSI-health, Anthropological index online, ISI. Ethnographies, textbooks and conference proceedings covering Indigenous themes were included and helped to identify leading authors in this field, who were contacted by phone or email. Six field trips of up to 2 weeks duration enabled the first author to establish relationships with key Indigenous and non-Indigenous informants and undertake participant observation of both formal 'ranger' programs and informal, customary management practices. While no previously validated measures of caring for country were identified, five potential questionnaire items were identified from extant literature: time on country, burning, using country, protecting country and ceremony. (Povinelli, 1993, Rose, 1992, p106–7).
Stage 2: Content Validity assessment with non-Indigenous informants
Four male non-Indigenous informants were identified during the scoping study. All had lived and worked in remote Indigenous communities for over 20 years. Three of these informants were still resident in remote communities at the time of consultation. One was resident in Darwin but maintained active involvement in Indigenous ranger programs. Three of the non-Indigenous informants had a direct association with the research community, and the fourth had no direct association with the community but was resident in a remote coastal Aboriginal community.
A sixth scale item, production of artefacts, was suggested and several plain language cues for each item were volunteered, corresponding to colloquial expressions in the community. An additional item concerning the reciprocal nature of caring for country, specifically the energy and vitality that arose from participation. (Thomson, 1975), was suggested by one informant. Subdivision of ceremonial activity between funeral rites and other ceremonies was also suggested
Stage 3: Content Validity assessment with Indigenous informants
Five Indigenous informants, four from the research community and one from outside the community assessed the content validity of the questionnaire. The four community informants (3 male and one female) were a purposive sample. All aged in their fifth decade, they were employees of disparate community agencies. They were from four different language groups. The community male informants represented the Indigenous rangers, an executive from the outstation resource centre and a member of the community health board. The community female informant was an employee of the women's centre. All community informants were fluent in English and several local Indigenous languages, identified with landowning groups and had in their lifetimes lived for extensive periods of time in remote homelands. The final Indigenous male informant, aged in his fourth decade, was based in Darwin and had over ten years of experience facilitating formal Indigenous natural resource management programs in widespread locations across the NT.
All Indigenous informants readily understood the purpose of the questionnaire and did not volunteer any additional items. The item regarding energy arising from caring for country was considered to be real and important but too difficult to include in the linguistically diverse research setting and was excluded at this stage. Item specific cues and quantification cues were considered intelligible and appropriate. The need for an interviewer administered questionnaire was highlighted. Three Indigenous informants felt that the division of ceremonial activity between funeral rites and other ceremonies was an artificial one and these two items were combined into a single ceremony category.
Stage 4: Construct validity assessment through key informant interview
Finally, a semi-structured interview with an Indigenous male from the community was undertaken based on the caring for country questionnaire developed in the previous three stages. This key informant, aged in his fifth decade, had well developed links across all language groups through his employment as an Aboriginal mental health worker. He had spent extensive periods in both homelands and in employment with non-Indigenous agencies in the township setting. This discussion was recorded on a digital voice recorder, predominantly in English, at the choice of the interviewee, but supplemented with Indigenous language to convey key concepts. Translations, where necessary, were supplied by the principal informant and verified .
In this community, caring for country activities were qualitatively associated with an holistic health construct, an-ngurrunga-wana, a state of vitality of mind, body and soul, roughly translated as "he-soul-big" . This construct forms the a priori hypothesis for measure development. We expected all scale items would load on a single latent factor, an-ngurrunga-wana.
Further construct validity assessment of the items within the questionnaire was also guided by Reid's  'Body, Land and Spirit' domains – an interpretive framework of Yolngu health beliefs in north east Arnhem land. Time on country, using country and burning are linked in practice , and involve direct interaction with specific landscapes. These three items may pertain to the dimension of land. Ceremony and protecting country are linked to spiritual beliefs and practices to maintain the spiritual integrity of landscapes  and may pertain to the dimension of spirit. The production of artefacts: carvings, paintings, weavings and other decorative or utility items are concrete expressions of specific landscapes or ancestral knowledge [35, 40]. Artefacts may thus pertain to body or the 'material embodiment' of land and spirit domains. (Additional file 2)
The interviewer also collected data on primary place of residence, education, income, diet, physical activity, alcohol consumption and smoking status. While we expected that participants engaging in higher levels of caring for country would come from homelands [25, 26], we wished to control for residence in our analysis because (i) caring for country participants could come from the township; (ii) not all homelands residents care for country and (iii) homelands residents may have differing dietary and physical activity factors, based on their isolation, which could potentially confound caring for country in predicting BMI.
Income was divided into three ordinal categories: 1 = unemployment benefits (lowest income); 2 = Abstudy (Aboriginal education support payments), Community Development Employment Program, carer allowance, child support, receiving payments for artefact production (middle income); 3 = salaried positions (highest income). This last category was rare, including only 1.3% of respondents. Educational attainment was categorised as: 1: no formal education; 2: primary education; 3: lower secondary; 4: year ten; 5: year twelve; 6: post school qualification. Higher levels of education have been asserted to deliver better health outcomes in Indigenous populations .
Diet data were collected with standardised visual cues depicting commonly available foodstuffs that participants reported consuming: never; sometimes; most days; every day. Physical activity was assessed by a question adapted from the Australian longitudinal study on women's health: "How many times a week do you exercise enough to get short of breath or huff and puff?" . This was accompanied by visual cues depicting sporting activity, hunting, digging and ceremonial dancing. Participants reported; none; one or two times; three or four times; more than four times. Smoking status was assessed by asking: "Do you smoke tobacco?" (yes/no).
Weight was recorded on digital scales to the nearest 100 g and height to the nearest centimetre, using a mounted stadiometer, following accepted techniques . Participants wore light clothing and had bare feet. BMI was derived by dividing weight in kilograms by the square of the person's height in metres.
Descriptive statistics were computed for the cohort. Item endorsement, inter-item correlations, quadratic weighted kappa scores for test-retest and proxy response reliability were calculated. Exploratory factor analysis was appropriate for a preliminary investigation of the factor structure underlying the items. The dataset was appropriate for exploratory factor analysis : the questionnaire items were theoretically related; the study was designed for factor analysis; the dataset was factorable (multiple inter-item correlations > .3); the sample size was adequate and; sampling statistics were acceptable (Kaiser-Meyer-Olkin statistic = .84, Bartlett's test of sphericity p < .0001). Maximum likelihood factoring with oblimin rotation were used to allow for skewed data and a correlated solution (should the solution contain more than one factor) respectively. Evaluation criteria were consistency with theory, factor loadings exceeding .45 and eigenvalues > 1.0.
One-factor congeneric modelling (see Berry ), a sub-set of structural equations modelling, was used to (i) test and refine the exploratory factor solution and (ii) generate a set of accurate item weightings for the creation of a weighted composite scale (Range = .76–3.06, M = 1.93, Sx = .67). The model was fitted using an asymptotic distribution free algorithm to accommodate non-normal distributions in the data. Model fit was assessed by a holistic appraisal of the χ2 statistic, critical ratios, and a selection of goodness of fit indices (absolute fit, incremental fit and parsimony indices).
Recent research has suggested that breadth of community participation across a range of important types of participation is more strongly linked to wellbeing than is the mean amount of participation, or very high levels of any particular type of participation . We investigated whether this may be true of any association between participating in caring for country and health. In our regression analyses, we compared a single weighted composite score for caring for country, derived from the one-factor model, with an index of total number of types of caring for country in which respondents participated. The index was created by dichotomising item scores by mean split, assigning a value of 0 (non-participator – below the mean) and 1 (participator – at or above the mean), and summing these scores. This generated a seven-point index (Range = 0–6, M = 2.73, Sx = 2.22). An un-weighted composite scale and the index demonstrated satisfactory internal coherence (Cronbach alpha scores = .88 and .85 respectively). Cronbach alpha scores cannot be derived for weighted composites.
Multivariate logistic and ordinal logistic regression analyses were performed to evaluate whether caring for country predicted obesity-related health behaviours, controlling for socio-demographic factors and other health behaviours. Multiple hierarchical regression models were used to test the relationship between caring for country and BMI. The models included variables tapping social determinants, residence, health behaviours and caring for country. Analyses were performed using Stata , SPSS , and AMOS (structural equation modelling) .