Sampling design is critical to the overall methodology of quantitative research studies, yet it does not always receive appropriate and timely attention in studies using nursing homes (NHs) as the unit of analysis. When samples are poorly designed, generalizability is limited and the validity of the analysis is questionable (Polit & Beck, 2012; Watson, Atkinson, & Egerton, 2006; Williamson, 2003). Although budgetary and time constraints can limit the overall size of a NH sample, careful planning can ensure that the sample is as representative as possible. In addition, it is often possible to test the representativeness of the obtained sample once data collection is complete and use weights to adjust for any observed differences. Using the example of a recently conducted study on the quality of life (QoL) of individuals with dementia living in Australian NHs (referred to as residential aged care facilities in Australia), the current article demonstrates the value of balancing probability techniques with practicalities in producing a sample as representative as possible given existing constraints. Recommendations are made for incorporating design elements to maximize the generalizability of a sample.
Probability sampling comprises a family of techniques in which every element of the population has a known, non-zero chance of being chosen for analysis (Polit & Beck, 2012; Tappen, 2011). In the example of a NH study for an entire country, this would mean that all NHs would have at least a small chance of being chosen for the sample. The most familiar of these techniques is the simple random sample (SRS), in which every element of the population is contained on a “list” from which the sample is drawn randomly. In the case of a SRS, every element has an equal chance of being chosen. The advantage of using probability sampling techniques is that the representativeness of the sample is maximized (Moore & McCabe, 2006).
Nonprobability sampling, given that it does not include a random element, rarely produces a representative sample. The most familiar sampling techniques in this category include:
- Convenience—the weakest of the designs, in which elements (e.g., NHs) are chosen from those close by, known to the investigators, or in another convenient manner;
- Purposive—in which elements are chosen to represent diversity on chosen characteristics, but not in a random manner;
- Quota—in which elements are chosen so that pre-defined distributions of characteristics are met (often referred to as “filling boxes”); and
- Consecutive—in which elements are chosen from a defined population over time. An example would be recruiting from all individuals entering an emergency department over a given week (Polit & Beck, 2012; Wood & Ross-Kerr, 2011).
Sampling Methods in Recent Quality of Life Studies
A nonsystematic literature search via Sciencedirect, on studies published between 2004 and 2014, using the search string “dementia” or “Alzheimer's Disease” AND “RACF” or “Nursing home” or “Residential care” AND “Quality of life” or “QoL,” generated 4,751 studies, yielding 41 on the QoL of individuals with dementia in NHs. More than one half of these studies (n = 23, 56%) did not provide enough information to fully evaluate the type and quality of the sampling design used. Yet, it was evident that a large proportion of studies (n = 35, 85%) used nonprobability designs; more than one half (n = 23, 56%) were convenience samples. Less than one third (n = 7, 30%) of studies that used convenience sampling noted it as a limitation of the study. Four studies used population sampling by attempting to recruit the whole population of NHs in a city or country. The only two studies that used probability sampling techniques were large studies, covering a state and country, respectively (Abrahamson et al., 2013; Vandervoort et al., 2013). Significantly, none of the reviewed studies tested the representativeness of their obtained sample against the population from which it was drawn.
The above example of literature could be drawn from any number of areas in nursing, psychology, or social sciences where convenience sampling and other nonprobability sampling techniques predominate (Gerrish & Lacey, 2010; Polit & Beck, 2012; Williamson, 2003). Recruitment in gerontological and geriatric research has previously been identified as problematic (Beattie, O'Reilly, Brooks, et al., 2015). Despite these recruitment difficulties, several large-scale QoL studies have been conducted on the regional (Beer et al., 2010; Chenoweth et al., 2011), state (Abrahamson et al., 2013; Zimmerman et al., 2005), national (Bouman et al., 2010; Vandervoort et al., 2013), and international (Verbeek et al., 2012) scale. Some good examples of probability sampling exist in the broader NH literature, which show variations in method, but all of which have strong representative designs (Hawes, Phillips, Rose, Holan, & Sherman, 2003; Jones, Dwyer, Bercovitz, & Strahan, 2009; Zimmerman, Gruber-Baldini, Hebel, Sloane, & Magaziner, 2002).
Underlying the quantitative arm of a large cross-sectional, mixed-methods study on the QoL of individuals with dementia in Australian NHs (Beattie, O'Reilly, Moyle, et al., 2015) was a carefully designed sampling methodology using the most robust techniques possible. Human research ethics clearance was obtained from the universities involved (i.e., Queensland University of Technology, Griffith University, University of Technology Sydney, University of Tasmania, LaTrobe University, and Curtin University) and from any NH or organization that required it.
Australia has a unique demographic distribution with a large geographic area of 4.8 million square miles and a relatively small population now approaching 24 million, combined with high concentrations in a small number of large cities. More than three fifths of Australia's total population lives in its largest five cities—which also serve as the capital cities of the most populous states in terms of people and NHs: Sydney, New South Wales (NSW); Melbourne, Victoria (VIC); Brisbane, Queensland; Perth, Western Australia; and Adelaide, South Australia (Table 1). As displayed in Figure 1, a five-tiered system is used by the Australian Bureau of Statistics (2006) to divide the country by increasing distance from major metropolitan areas: major city, inner regional, outer regional, remote, and very remote areas. Remote and very remote areas in the interior cover much of Australia's geographic area, but contain very few people and only a small number of NHs (Table 1).
Area, Population, and Number of Nursing Home Facilities by State and Geographical Type
Remoteness classifications in Australia. Adapted from Australian Bureau of Statistics. (2014). 2006 ASGC remoteness structure. Retrieved from http://www.abs.gov.au/websitedbs/D3310114.nsf/home/remoteness+structure
The most significant aspect of the sampling methodology was determining which specific NHs would be approached for recruitment. The goal was to produce a sample that was diversified and as representative as possible of all NHs in Australia. Fortunately, the Australian government maintains an up-to-date list of all NHs in the country (N = 2,774); this list formed the study population. Although a full probability sample from this population would have been ideal, as with all research projects, constraints existed that prevented that approach. The following description of the sampling design process describes how these constraints were managed, in order (Table 2).
Nursing Homes Excluded From the Sample
Statistical. Anticipating a hierarchical analysis in which data at several different levels (i.e., NH, staff, and resident) would be used, it was determined that the optimal number of residents with dementia to be recruited per NH would be between five and 15. Given this, it was decided that NHs with <25 residents overall would be excluded from the potential sample. Under this constraint, 238 small NHs were excluded.
Logistical. Remote and very remote areas are difficult and expensive to reach physically for on-site data collection. Only a small percentage of NHs with >25 residents (n = 19, 2%) are in these areas and were also excluded from the sample.
Budgetary/Practical. Any study attempting national representation must consider how best to achieve collecting data in a number of areas. The current team was a collaborative group with membership covering six different universities from five different Australian states. This approach made collecting data in seven of Australia's eight states and territories possible. One extra territory, the Australian Capital Territory (ACT), and one extra state were covered by existing teams. The final remaining area, the Northern Territory (NT), is comprised primarily of very small, remote area NHs and was considered too difficult to include in the sample. The three remaining NT nursing homes were excluded from the sample.
In excluding some NHs from the sample in these three ways, they had zero chance of being chosen for the sample, thereby breaking the primary rule of probability sampling. However, as shown in Table 2, these NHs form a small percentage (n = 260, 9%) of the overall number of Australian NHs and contain an even smaller percentage (3%) of the total residential places available.
After removing the NHs described above, 2,514 remained to form the “sampling frame” (Figure 2); a probability technique was then used to select the final sample for recruitment. Disproportionate stratified random sampling suited the needs of the current study because of the desire to ensure that types of NHs with small numbers (e.g., those in outer regional areas) were represented. Stratified designs involve dividing the total population into strata defined by a variable on which the researcher wishes to ensure representation. In the current study, three stages of stratification were used, with defined numbers of NHs to be recruited for each category.
Sampling methodology. Note. RACFs = residential aged care facilities.
In the current study, the first stratum was state; this division was primarily to break the recruiting into reasonable sizes for each research group. The second stratum was geographical type, with more NHs to be sampled in the more urban areas (given their higher population concentrations). The third stratum was organization type, dividing NHs into charitable/religious, public/community, and private (i.e., for profit) types. Table 3 and Figure 2 show the complete list of all Australian NHs within each stratum. Once divided into these strata, the resulting “lists” of NHs were randomly sorted using the RANDOM function in SPSS version 21. Research teams in each state were instructed to recruit from each list pertaining to their assigned state/territory, beginning from the top and working down, until the desired number of NHs was recruited. No changes to the list were permitted and research teams were not allowed to eliminate NHs from the list that they considered difficult to recruit.
Achieved Nursing Home Sample Across Stratification Variables and Goodness of Fit With Population of Australian Nursing Homes
The final sample included 53 NHs with the desired balance across the three strata (Table 3). In evaluating the representativeness of the achieved sample relative to the total population of NHs in Australia, chi-square goodness of fit tests were conducted for each of the stratum used in the sampling design. A significant value of ≤0.05 on this measure would indicate that the distribution of the variable tested in the achieved sample was different from that in the study population.
The sample size by state was explicitly designed in such a way as to (a) exclude facilities in the NT, (b) over-sample the smaller population states of Tasmania and ACT, and (c) under-sample the largest states of NSW and VIC. The distribution of NHs in the achieved sample by state and territory was therefore skewed as substantiated by the significant chi-square test (<0.001).
By design, no NHs from remote and very remote geographic classifications were included in the study population; only a small percentage (2.1%) of all Australian NHs were in these areas. Subsequently, the chi-square test was not significant.
Unlike the other two stratifying variables, no categories of the organizational type variable were excluded from the sampling frame. The distribution across this variable in the population and the achieved sample were similar, and the chi-square value was not significant.
The degree of sample representativeness was tested by comparing the distribution of the achieved sample across the variables used in the stratification process with the distribution for all Australian NHs. The appropriate statistical test in this case is the chi-square goodness of fit test, which is calculated differently from the more familiar chi-square test of association (Dietz & Kalof, 2009). In the current study, the achieved sample was found to be adequately representative by organizational and geographical type, but not by state/territory. The state result was expected given the design and widely differing distribution of population by state and territory in Australia (Table 1). The design ensured that the smaller population states had adequate sample sizes to produce reliable results in a statistical sense.
If desired for statistical or substantive reasons, sample weights can be calculated to “correct” for lack of representativeness of a population (Smyth et al., 2013; Tracy & Carkin, 2014). This process requires that either (a) the (approximate) distribution of the population is known or (b) the exact percentages used in the sampling design for each stratum are known. For the sake of simplicity, the calculation of weights in the current case is discussed as if there were only one stratification variable. In the first scenario, weights are calculated using the inverse of division of the percentage in the sample by the percentage in the population. In the second scenario, the weight is the inverse of the division of the design percentage over the population percentage (Tracy & Carkin, 2014). Once these weights are applied, the distribution of the sample achieved over the relevant variable will closely resemble that of the population. At the analysis stage, the normal procedure is to run all analyses on the unweighted data and then on the weighted data. If results do not vary significantly, unweighted results should be used.
In the strictest interpretation of the current study's achieved sample, it represents all Australian NHs except those that are very small, geographically remote, or in the NT. These restrictions would need to be admitted and discussed in any article resulting from the current research study. The largest class of NHs excluded were those with <25 beds, of which there are a significant number in Australia. Every conclusion about QoL of individuals with dementia resulting from the current study would therefore need to be qualified by saying it refers only to those living in NHs with ≥25 beds. The value of using probability sampling for the current study is in producing the best possible representation of the population, given existing constraints.
Recommendations for Conducting a Probability-Based Study
The recommendation of applying a probability-based sampling strategy may seem daunting or overly complex, but the process is straightforward when some forethought and careful planning are applied. The result will be a more statistically robust study. Although the current study concerned QoL in NHs and was complex in the sense that it used three stratifying variables, probability sampling techniques could be applied to any topic under investigation in gerontological nursing with the goal of obtaining a representative sample for any population that is well described. There are four highly used types of probability sampling designs, all of which meet the required criterion that every element of the population has a non-zero chance of being included in the sample (Moore & McCabe, 2006; Ott & Longnecker, 2010; Polit & Beck, 2012):
- Simple random—as previously described, this technique involves sampling randomly from a full list of the study population.
- Systematic—sampling every nth element of the study population (note that this is a probability design only if the starting value is random, often referred to as a “random seed”). For this type, an alphabetic sorting by last name of individuals is not recommended.
- Stratified—as illustrated in the current article, this is a technique to ensure smaller groups in the population are represented in the sample.
- Cluster—with this technique, sampling is conducted at two (or more) levels (e.g., by state, county, then NH), with random or systematic sampling at each level. Cluster sampling can be considerably more cost-efficient than other probability techniques, but weighting and calculation of appropriate standard errors can be more complicated.
In thinking about what type of sampling methodology should or can be applied for a given study, it is helpful to start by considering the following questions:
- What, or who, is trying to be “represented” in the study? In other words, what is the scope of the study? For example, a sample representing a particular group or groups may be the intended achievement, in which case the study need only start from that point, identify the specific study population in any dissemination of the research, and consider any limitations that could impose on generalizability of the results to other populations (Ott & Longnecker, 2010).
- Is a “list” available for my population of interest? In the best case scenario, as described in the current article, a full list of all members of the population is available, allowing for more freedom in designing the sampling methodology (Moore & McCabe, 2006).
- What constraints (budgetary, statistical, logistical, or practical) apply to the research situation? How do these constraints affect what can be done in terms of sampling design and recruitment?
- Are there groups/types within the population that are important to ensure representation of in the sample (Ott & Longnecker, 2010)?
In considering the scope of the project, the main recommendation is to be clear from the conception of the study exactly who is intended to be the population, and if it is limited in some way, have an explicit justification. For example, if only interested in understanding staff turnover in large NHs, the study population could be limited to those with >50 beds.
If unable to access a full sampling frame (i.e., list), a probability design can be achieved through cluster sampling. Alternatively, two nonprobability techniques that could achieve a reasonable approximation of a probability sample if done well are quota and purposive sampling. If nonprobability samples are used, it is important to take into account any potential bias in the results and discuss these as limitations to the study.
As discussed in the example in the current article, acknowledging and deciding how to manage constraints early on in a research project will allow researchers to better describe and test the representativeness of the sample once achieved. To better represent a large geographical area, such as a region or entire country, collaboration with other research groups, universities, or organizations can solve budgetary and logistical issues and move research in a field forward. In the current study, only one research group needed to travel for multiple nights to a different state to conduct data collection, with all other data collection handled within-state by existing collaborators. Once all constraints are managed and any nonresearchable population is excluded, the remaining population becomes the sampling frame and defines the population that will be represented by the study.
Populations often contain specific groups or types of organizations that are of interest, but which are not represented in high numbers (e.g., specific cultural or racial groups, rural NHs). Stratified or clustered sampling designs, with over-weighting of small groups, are recommended in these cases to ensure that these groups are represented in the final sample (Ott & Longnecker, 2010; Polit & Beck, 2012). Stratified designs can be performed with either defined numbers or percentages in each category, or simply using an overall percentage for each stratum. As described above, sample weights can be applied if one of these types of sampling methods affects overall conclusions about the population.
Sampling design, despite being crucial to the generalizability, validity, and reliability of research results, is often poorly planned, and potential avenues for the application of probabilistic methodology are not considered. The nursing field is not alone in this respect, with similar problems common across the broader medical field and psychology (Williamson, 2003). In particular, the use of convenience samples has the potential to introduce serious potential biases into research results. For all studies, even those with probability designs, it is crucial to test the representativeness of the achieved sample against any information known about the distribution of relevant variables in the study population. The current article provides recommendations for considering sampling design issues throughout a project, from conception through data analysis, so as to produce more robust results.
- Abrahamson, K., Lewis, T., Perkins, A., Clark, D., Nazir, A. & Arling, G. (2013). The influence of cognitive impairment, special care unit placement and nursing facility characteristics on resident quality of life. Journal of Aging and Health, 25, 574–588. doi:10.1177/0898264313480240 [CrossRef]
- Australian Bureau of Statistics. (2006). Australian Standard Geo graphical Classification (ASGC) Statistical Geography (Vol. 1). Canberra, Australia: Author.
- Australian Bureau of Statistics. (2011). 2011 census of population and housing. Retrieved from http://www.abs.gov.au/websitedbs/censushome.nsf/home/tablebuilder?opendocument&navpos=240
- Australian Bureau of Statistics. (2014). 2006 ASGC remoteness structure. Retrieved from http://www.abs.gov.au/websitedbs/D3310114.nsf/home/remoteness+structure
- Australian Bureau of Statistics. (2015). 2011 census of population and housing. Retrieved from http://www.abs.gov.au/websitedbs/censushome.nsf/home/data
- Beattie, E., O'Reilly, M., Brooks, D., MacAndrew, M., McCrow, J. & Fielding, E. (2015). Research recruitment of people with dementia and their carers: Lessons learned from nine studies. Manuscript submitted for publication.
- Beattie, E., O'Reilly, M., Moyle, W., Chenoweth, L., Fetherstonhaugh, D., Horner, B. & Fielding, E. (2015). Multiple perspectives on quality of life for residents with dementia in long term care facilities: Protocol for a comprehensive Australian study. International Psychogeriatrics, 27, 1739–1747. doi:10.1017/S1041610215000435 [CrossRef]
- Beer, C.D., Horner, B., Almeida, O.P., Scherer, S., Lautenschlager, N.T., Bretland, N. & Flicker, L. (2010). Study protocol dementia in residential care: Education intervention trial (DIRECT); Protocol for a randomised controlled trial. Trials, 11, 1–6. doi:10.1186/1745-6215-11-63 [CrossRef]
- Bouman, A.I.E., Ettema, T.P., Wetzels, R.B., van Beek, A.P.A., de Lange, J. & Droes, R.M. (2010). Evaluation of Qualidem: A dementia-specific quality of life instrument for persons with dementia in residential settings; scalability and reliability of subscales in four Dutch field surveys. International Journal of Geriatric Psychiatry, 26, 711–722. doi:10.1002/gps.2585 [CrossRef]
- Chenoweth, L., King, M., Luscombe, G., Forbes, I., Jeon, Y., Stein-Parbury, J. & Haas, M. (2011). Study protocol of a randomised controlled group trial of client and care outcomes in the residential dementia care setting. Worldviews on Evidence-Based Nursing, 8, 153–165. doi:10.1111/j.1741-6787.2010.00204.x [CrossRef]
- Dietz, T. & Kalof, L. (2009). Introduction to social statistics: The logic of statistical reasoning. New York, NY: Wiley-Blackwell.
- Geoscience Australia. (2004). Land areas of states and territories. Retrieved from http://www.ga.gov.au/scientific-topics/geographic-information/dimensions/area-of-australia-states-and-territories
- Gerrish, K. & Lacey, A. (Eds.). (2010). The research process in nursing (Vol. 6). West Sussex, UK: Wiley-Blackwell.
- Hawes, C., Phillips, C.D., Rose, M., Holan, S. & Sherman, M. (2003). A national survey of assisted living facilities. The Gerontologist, 43, 875–882. doi:10.1093/geront/43.6.875 [CrossRef]
- Jones, A.L., Dwyer, L.L., Bercovitz, A.R. & Strahan, G.W. (2009). The National Nursing Home Survey: 2004 overview. Vital and Health Statistics, 13, 1–155.
- Moore, D.S. & McCabe, G.P. (2006). Introduction to the practice of statistics. New York, NY: W.H. Freeman and Company.
- Ott, R.L. & Longnecker, M. (2010). An introduction to statistical methods of data analysis (6th ed.). Belmont, CA: Brooks/Cole.
- Polit, D.F. & Beck, T.B. (2012). Nursing research: Generating and assessing evidence for nursing practice (9th ed.). Philadelphia, PA: Wolters Kluwer.
- Smyth, W., Fielding, E., Beattie, E., Gardner, A., Moyle, W., Franklin, S. & MacAndrew, M. (2013). A survey-based study of knowledge of Alzheimer's disease among health care staff. BMC Geriatrics, 13, 1–8. doi:10.1186/1471-2318-13-2 [CrossRef]
- Tappen, R.M. (2011). Advanced nursing research: From theory to practice. Sudbury, MA: Jones and Bartlett Learning.
- Tracy, P.E. & Carkin, D.M. (2014). Adjusting for design effects in disproportionate stratified sampling designs through weighting. Crime and Delinquency, 60, 306–325. doi:10.1177/0011128714522114 [CrossRef]
- Vandervoort, A., Van den Block, L., Van der Steen, J.T., Volicer, L., Stichele, R.V., Houttekier, D. & Deliens, L. (2013). Nursing home residents dying with dementia in Flanders, Belgium: A nationwide postmortem study on clinical characteristics and quality of dying. Journal of the American Medical Directors Association, 14, 485–492. doi:10.1016/j.jamda.2013.01.016 [CrossRef]
- Verbeek, H., Meyer, G., Leino-Kilpi, H., Zabalegui, A., Hallberg, I.R., Saks, K. & Hamers, J.P. (2012). A European study investigating patterns of transition from home care towards institutional dementia care: The protocol of a RightTimePlaceCare study. BMC Public Health, 12, 68. doi:10.1186/1471-2458-12-68 [CrossRef]
- Watson, R., Atkinson, I. & Egerton, P. (2006). Successful statistics for nursing and healthcare. New York, NY: Palgrave Macmillan.
- Williamson, G.R. (2003). Misrepresenting random sampling? A systematic review of research papers in the Journal of Advanced Nursing. Journal of Advanced Nursing, 44, 278–288. doi:10.1046/j.1365-2648.2003.02803.x [CrossRef]
- Wood, M.J. & Ross-Kerr, J.C. (2011). Basic steps in planning nursing research: From question to proposal (7th ed.). Sudbury, MA: Jones and Bartlett.
- Zimmerman, S., Gruber-Baldini, A.L., Hebel, J.R., Sloane, P.D. & Magaziner, J. (2002). Nursing home facility risk factors for infection and hospitalization: Importance of registered nurse turnover, administration, and social factors. Journal of the American Geriatrics Society, 50, 1987–1995. doi:10.1046/j.1532-5415.2002.50610.x [CrossRef]
- Zimmerman, S., Sloane, P.D., Williams, C.S., Reed, P.S., Preisser, J.S., Eckert, J.K. & Dobbs, D. (2005). Dementia care and quality of life in assisted living and nursing homes. The Gerontologist, 45, 133–146. doi:10.1093/geront/45.suppl_1.133 [CrossRef]
Area, Population, and Number of Nursing Home Facilities by State and Geographical Type
|Variable||Geographical Area (mi2)||Population (2011), n (%)||Facilities, n (%)|
| New South Wales||497,496||6,917,656 (32.2)||887 (32)|
| Victoria||141,310||5,354,040 (24.9)||774 (27.9)|
| Queensland||1,075,375||4,332,737 (20.1)||480 (17.3)|
| Western Australia||1,571,991||2,239,171 (10.4)||244 (8.8)|
| South Australia||611,107||1,596,570 (7.4)||267 (9.6)|
| Tasmania||42,502||495,351 (2.3)||82 (3)|
| Australian Capital Territory||1,465||357,218 (1.7)||26 (0.9)|
| Northern Territory||838,310||211,943 (1)||14 (0.5)|
|Total||4,779,556a||21,504,686b (100)||2,774 (100)|
| Major city||15,006,518 (69.8)||1,694 (61.1)|
| Inner regional||3,998,045 (18.6)||690 (24.9)|
| Outer regional||1,963,407 (9.1)||332 (12)|
| Remote||300,105 (1.4)||36 (1.3)|
| Very remote||187,639 (0.9)||22 (0.8)|
| Other||49,003 (0.2)||—|
|Total||21,504,717c (100)||2,774 (100)|
Nursing Homes Excluded From the Sample
|Type of Nursing Home||Facilities, n (%)||Places, n (%)|
|Total||2,774 (100)||179,749 (100)|
| Very small (<25 residents)||238 (8.5)||3,867 (2.2)|
| Remote/very remote||19 (0.7)||864 (0.5)|
| Northern Territory||3 (0.1)||250 (0.1)|
|Remaining in study population||2,514 (90.6)||174,768 (97.2)|
Achieved Nursing Home Sample Across Stratification Variables and Goodness of Fit With Population of Australian Nursing Homes
|Strata||Population, n (%)||Sample, n (%)||Chi-Square||df||p Value|
| New South Wales||887 (32)||10 (18.9)||31||7||<0.001|
| Victoria||774 (27.9)||10 (18.9)|
| Queensland||480 (17.3)||10 (18.9)|
| Western Australia||244 (8.8)||10 (18.9)|
| South Australia||267 (9.6)||5 (9.4)|
| Tasmania||82 (3)||5 (9.4)|
| Australian Capital Territory||26 (0.9)||3 (5.7)|
| Northern Territory||14 (0.5)||—|
| Major city||1,694 (61.1)||26 (49.1)||6.38||4||0.172|
| Inner regional||690 (24.9)||16 (30.2)|
| Outer regional||332 (12)||11 (20.8)|
| Remote||36 (1.3)||—|
| Very remote||22 (0.8)||—|
| Religious/charitable||1,213 (43.7)||23 (43.4)||0.03||2||0.984|
| Government||756 (27.3)||15 (28.3)|
| Private||805 (29)||15 (28.3)|