Methodology

BludgerTrack 2019 is an aggregate of all published federal opinion polls, adjusted to account for observed biases and weighted according past reliability. Local regression analyis (LOESS) is used to plot trendlines through the available data and determine current results based on the modelled results for the most recent available point in time. State breakdowns from the polling, both published and unpublished, are used to track each state's deviation from the national result. These are added together to calculate state-level voting intention results, which are used to project seat totals.

Primary vote results for the Coalition, Labor and the Greens are subject to each pollster's bias adjustments. The preferred measure for the adjustments is an average of errors in final pre-election polls for the 2016 federal election and, as far as possible, the most recent elections for each of the six states. In calculating these averages, each election is weighted by its voting population. This method is used for Newspoll, Galaxy and Essential Research.

For the pollsters with insufficient historical data, ReachTEL, Ipsos and YouGov/Fifty Acres, the comparison is made with a trend measure of polling from the other three pollsters through the current parliamentary term. Bias measures are halved to account for pollsters' tendency to adjust methodology in response to past failure.

In determining the trend measures, a weighting measure is applied for each pollster based on its accuracy and frequency of publication. The accuracy measure reduces the weighting of pollsters with larger bias adjustments, and equals one divided by an average of the pollster's bias adjustments for the three parties, weighted according to their share of the total vote. This measure is then divided by the the number of the pollster's polls in the model, to limit the influence of frequently reporting pollsters. However, the value for Newspoll is used as a ceiling that no other pollster exceeds. These measures are variable, but were as shown at right as of May 2018. Also displayed are the bias adjustments before being halved.

Two-party results are projected for each electorate based on the state swing figures, subject for adjustment according to candidate factors. The following measures, which have been determined through regression analysis of past election results, are applied cumulatively to the incumbent party's two-party vote. For the mainland states, a small compensating adjustment is made to seats where none of the factors below applies, so that the overall state result remains consistent with the state poll trend.

The adjusted two-party results for each seat are used to determine Coalition win probabilities using a normal distribution function. Historical observation has been used to determine a relationship of y=0.131x+0.019 between the standard deviation (y) of swings within a given state and the state's overall swing (x). This works out to a 1.9% standard deviation in the event of no swing, increasing to 3.2% for a swing of 10%.

Since BludgerTrack is essentially a two-party model, complications are presented by the five seats not won by the major parties in 2016: Kennedy (Bob Katter, Independent, Queensland), Denison (Andrew Wilkie, Independent, Tasmania), Melbourne (Adam Bandt, Greens, Victoria), Indi (Cathy McGowan, Independent, Victoria) and Mayo (Rebekha Sharkie, Nick Xenophon Team, South Australia). Absent electorate-level poll results to the contrary, it will be assumed these seats will be retained by their incumbents, and that no further seats will be gained by minor parties or independents.