Abstract

Addendum to:

Biddle, N., Ellen, L., Korda, R., and Reddy, K., (2020) ‘Suicide Mortality in Australia: Estimating and Projecting Monthly Variation and Trends From 2007 to 2018 and Beyond’ Suicide Mortality in Australia: Estimating and Projecting Monthly Variation and Trends From 2007 to 2018 and Beyond | ANU Centre for Social Research & Methods

This addendum provides a descriptive comparison of the monthly forecasted deaths for the period of January 2019 till November 2019 from Biddle, Ellen, Korda and Reddy (henceforth, BEKR 2020), with preliminary suicide deaths data for this same period, adjusted for the expected revisions process. The final forecasting model used within BEKR (2020) was an Exponential State Smoothing Model (ETS) Model. Five different models were initially trained with the ETS model selected because of its higher comparative accuracy (measured using mean absolute percentage error) and after a visual inspection of residual diagnostics. The ETS modelled forecasts have been used for descriptive comparison with the actual preliminary suicide deaths data. However, it is noted that the ETS model performance was only marginally superior to the Exponential Smoothing State Space Model with Box-Cox Transformation (TBATS) model also trained. The ETS modelled death forecasts provided within BEKR (2020) are designed to estimate Final suicide deaths data. Currently the most updated 2019 suicide deaths data is Preliminary.1 To allow for accurate comparisons, and based on historic estimates of the difference between the preliminary and final estimates, the deaths figures are multiplied by 1.133. The month of December was excluded from the comparison of forecasted and actual preliminary deaths data because preliminary deaths data for December appears to be differentially impacted by the data revisions process when compared to other calendar months. This may be explained by a larger number of deaths occurring in December not being officially registered until the following calendar year. Figure 1 shows the monthly point forecasted death rates alongside the actual (adjusted) death rates calculated using the preliminary deaths data provided by the AIHW. Deaths rates per 100 000 population were calculated using the method outlined within oursubstantive report. Figure 2 provides that same information as Figure 1, with the addition of confidence intervals around monthly point forecasts. Table 1 presents the modeled death rates per 100 000 population and the actual (adjusted) Preliminary death rates and counts

File attachments