The response rates for each of the profession surveys are listed in Table 1.
Table 1: Survey response rates, states and territories, 2013 | NSW | Vic | Qld | WA | SA | Tas | ACT | NT | Australia(a) | Psychologists | 85.8 | 82.8 | 83.4 | 81.9 | 85.2 | 86.0 | 84.2 | 86.9 | 84.2 | Pharmacists | 86.1 | 87.6 | 85.8 | 87.9 | 85.1 | 83.8 | 88.8 | 85.8 | 86.6 | Physiotherapists | 92.7 | 91.8 | 91.2 | 90.6 | 89.5 | 95.5 | 95.9 | 93.2 | 91.8 | Dentists | 91.4 | 90.1 | 89.2 | 89.2 | 88.8 | 89.9 | 94.1 | 92.4 | 90.0 | Dental hygienists | 93.0 | 94.0 | 95.5 | 95.8 | 94.2 | 100.0 | 94.3 | 85.7 | 94.4 | Dental prosthetists | 93.3 | 86.6 | 91.0 | 94.3 | 94.3 | 100.0 | 93.3 | 100.0 | 91.3 | Dental therapists | 97.8 | 95.9 | 98.0 | 95.9 | 97.9 | 98.0 | 100.0 | 100.0 | 97.1 | Oral health therapists | 79.8 | 81.1 | 91.9 | 38.6 | 73.6 | 70.0 | 88.2 | 100.0 | 81.4 | Occupational therapists | 88.7 | 87.2 | 87.8 | 87.3 | 86.5 | 90.0 | 93.7 | 92.1 | 87.8 | Medical radiation practitioners | 84.7 | 89.7 | 86.7 | 86.8 | 86.0 | 92.4 | 89.6 | 82.2 | 86.8 | Chiropractors | 93.0 | 91.3 | 93.9 | 90.6 | 95.6 | 92.2 | 97.0 | 100.0 | 92.6 | Optometrists | 94.4 | 93.8 | 93.2 | 94.0 | 93.7 | 95.5 | 94.9 | 97.0 | 94.0 | Chinese medicine practitioners | 88.4 | 91.5 | 88.4 | 86.7 | 88.8 | 91.2 | 91.9 | 81.8 | 89.1 | Podiatrists | 92.6 | 90.8 | 91.6 | 89.9 | 93.4 | 93.0 | 96.2 | 87.5 | 91.6 | Osteopaths | 92.3 | 88.9 | 88.4 | 94.3 | 100.0 | 89.5 | 81.8 | 100.0 | 90.3 | Aboriginal and Torres Strait
Islander health practitioners. | 70.4 | 62.5 | 70.6 | 69.2 | 44.4 | 100.0 | 100.0 | 69.4 | 69.0 |
(a) Includes health workers who did not state or adequately describe their state or territory, and those who were overseas.
Source: AIHW NHWDS 2013.
Data are reported on the basis of the most current address at the time the survey was undertaken, unless stated otherwise. The data include employed health practitioners who did not state or adequately describe their location, as well as employed health practitioners who were overseas. The national estimates include these groups.
Estimation procedures
The AIHW uses registration data together with survey data to derive estimates of the total health practitioner workforce. Not all practitioners who receive a survey respond, because it is not mandatory to do so. In deriving the estimates, two sources of non-response to the survey are accounted for:
• item non-response—occurs as some respondents return partially completed surveys. Some survey records were so incomplete that it was decided to omit them from the reported survey data.
• survey non-response—occurs because not all registered medical practitioners who receive a questionnaire respond.
Imputation methods are used account for item non-response and survey non-response.
Imputation: estimation for item non-response
The imputation process involves an initial examination of all information provided by a respondent. If possible, a reasonable assumption is made about any missing information based on responses to other survey questions. For example, if a respondent provides information on hours worked and the area in which they work, but leaves the workforce question blank, it is reasonable to assume that they were employed.
Missing values remaining after this process are considered for their suitability for further imputation. Suitability is based on the level of non-response to that item.
In imputation, the known probabilities of particular responses occurring are used to assign a response to each record. Imputed values are based on the distribution of responses occurring in the responding sample. Therefore, fundamental to imputing missing values for survey respondents who returned partially completed questionnaires is the assumption that respondents who answer various questions are similar to those who do not.
Age values within each state and territory of principal practice are first imputed to account for missing values. Other variables deemed suitable for this process were then imputed. These included hours worked in the week before the survey, principal role of main job, principal area of main job in medicine and work setting of main job.
Imputation: estimation for population non-response
In 2013, the methodology for population non-response was changed from a weighting-based methodology to a hot deck-based imputation, similar to that used for imputing unreported hours in previous years.
The data were sorted into strata, so imputations were made using survey data from records that have similar registration details. The strata used for imputation were registration type (with limited registrants grouped together and specialist registrants grouped with those who also had general registration), a derived primary specialty categorisation, sex, age group, remoteness area and state, in that order.
Donor records were spaced evenly within strata to ensure records were used within the strata an equal number of times plus or minus 1, and that most strata within the hot deck were restricted to within stratum imputations. For example, if there were 5 respondents and 12 non-respondents in a cell, the expected number of uses would be 2.4, resulting in each donor being used either 2 or 3 times.
Because the data were imputed and not weighted, some data may be affected in different ways from those previously published. For example, because a practitioner’s location of main job is most likely to be the same as their registration address, this has been used for the location estimation of non-respondents. Using this estimate rather than weighting will improve the accuracy of estimates for small geographic areas, as previously weighted data would scale up data for individuals across the state/territory and the registration information for records would not be taken into account.
For variables not used in the imputation (that is, all variables other than the registration type, remoteness area, state and territory of principal practice, age and sex), it is assumed, for estimation purposes, that respondents and non-respondents have similar characteristics. If the assumption is incorrect, and non-respondents are different from respondents, then the estimates will have some bias. The extent of this cannot be measured without obtaining more detailed information about non-respondents. |