About

Start your brilliant career with a degree from Australia's #1 ranked university.

News

Find contact details for any general enquiries.

Engage with us

Find contact details for any general enquiries.

Study with us

Our graduates gain the knowledge and skills to lead organisations, develop public policy, create new companies and undertake research.

Our research

Our research is focused on issues that are highly significant for organisations, the Australian economy, and society at large.

Current Students

Whether you're a new or continuing student, you can find everything you need here about managing your program and the opportunities available to you.

Alumni

Our alumni may be found in the world’s leading companies, policy agencies and universities.

Contact us

Find contact details for any general enquiries.

Staff directory

Professor Alan Welsh

Professor Alan Welsh

Research School of Finance, Actuarial Studies & Statistics

B.Sc. Mathematical Statistics (Honours), Sydney 1982
Ph.D. Statistics, The Australian National University 1985

Position: Professor
Phone: +61 2 612 57313
Office: Room 3.68, CBE Bld (26C)

Magnitude-based Inference

Over the last decade, a method for analysing data called 'magnitude-based inference' has been developed and promoted in sport science as a new, improved statistical method. It has so far attracted little scrutiny by statisticians. This project considers 'magnitude-based inference' and its interpretation by examining in detail its use in the problem of comparing two means. The methodology is extracted from the spreadsheets which are provided to users of the analysis (Sport Science Page).  The implemented version of the method is compared with general descriptions of it so that the method can be interpreted in familiar statistical terms.
 
'Magnitude-based inference' is not a progressive improvement on modern statistics. It does not replace the use of p-values by direct inference about magnitudes (as in using confidence intervals) but rather uses two different probabities.  These probabilities are not directly related to confidence intervals but rather are interpretable either as p-values for two different, nonstandard tests (for different null hypotheses) or as approximate Bayesian calculations which also lead to a type of test.  This test is like using a standard test but at a very high level.  This explains both how the method is 'less conservative' than the standard test and why it is not a real improvement on that test.  The substantial reduction in sample sizes claimed for the method (30% of the sample size obtained from standard frequentist calculations) is not justifiable so the sample size calculations should not be used. Rather than use 'magnitude-based inference', a better solution is to be realistic about the limitations of the data and use either confidence intervals or a fully Bayesian analysis.
 

Paper in Medicine and Science in sport and Exercise (MSSE).

Slides from talk presented at AIS, Friday 22 August 2014

Animation 1:  Constructing the ternary diagram to interpret and show the effect of changing the thresholds  ηb and ηh

The ternary plot to represent pb, ph and 1−pb−ph is constructed by drawing a triangle.  Each probability is assigned to a vertex of the triangle.  We draw in the pb axis from the pb vertex to the centre of the opposite side.  The pb vertex represents pb=1, ph=0 and 1−pb−ph=0; the opposite side represents pb=0 and the values of ph and 1−ph. We draw in a line parallel to the opposite side at pb=0.05 and label it.  We repeat this for three additional values of pb.  Then we draw in the ph axis and the lines parallel to the side opposite the ph vertex to show the same 4 values of ph. We complete the ternary diagram by drawing in the 1−pb−ph axis and lines parallel to the base to show 4 values of 1−pb−ph.  These gridlines are not labelled to prevent visual clutter.

For 'magnitude-based inference', we draw in a threshold value for pb given by  ηb=0.25.  This threshold value partitions the triangle into 2 regions: a beneficial region (pb≥ηb) which we shade in blue and a not-beneficial region (pb<ηb) that we shade in grey.  We then draw in a threshold value for ph given by ηh=0.25.  The new threshold value partitions both the beneficial and the not-beneficial regions into 2 further regions.  First, the grey not-beneficial region is partitioned into a trivial region (pb<ηb and ph<ηh) which we leave shaded in grey and a harmful region (pb<ηb and ph≥ηh) that we shade in red.  Second, the blue beneficial region is partitioned into a beneficial region (pb≥ηb and ph<ηh) which we leave shaded in blue and an unclear region (pb≥ηb and ph≥ηh) that we shade in purple.

The animation then decreases the two thresholds together in a sequence of steps from ηb=ηh=0.25 to ηb=ηh=0.05 to present the effect of changing the threshold in 'mechanistic magnitude-based inference'.  Finally, the animation holds ηh=0.05 fixed and increases ηb in a sequence of steps from ηb=0.05 to ηb=0.25 to present the effect of changing the ηb threshold in 'clinical magnitude-based inference'.  

Animation 2: The effect of changing δ; on pb and ph in the ternary diagram and the probabilities of finding an effect when there is none 

The one-sided p-value for a single sample is shown on the base of the triangle.  The animation shows the path the triple pb, ph and 1−pb−ph traces through the ternary diagram as δ increases. This moves through the unclear to the beneficial and finally into the trivial region.

The animation is repeated for six different samples. The same pattern holds for samples with initial pb sufficiently large (p/2 sufficiently small); once the initial pb is smaller than 0.75, the path moves either through the unclear to the harmful and finally into the trivial region or through the harmful into the trivial region.

The next animation follows the triples pb, ph and 1−pb−ph for 200 random samples generated under the null hypothesis of no effect as δ increases.  The empirical distribution is curved in the simplex and all the points eventually move up into the trivial region.  The next sequence shows the same animation with plots of the empirical probabilities (based on 10,000 samples) of pb, ph and 1−pb−ph falling in the beneficial, trivial and harmful regions as δ increases.  The horizontal lines correspond to probabilities of 0.05, 0.25, 0.75 and 0.95; the dashed vertical line corresponds to δ=4.41, a recommended default value of δ

for this example.  Finally, we show larger versions of these empirical probability plots to show the details.

Animation 3:  The effect of changing δ on pb and ph, showing both the Frequentist and the Bayesian interpretations of these probabilities

The animation starts by showing the sampling distribution and the p-value for testing the null hypothesis of no effect.  The figure shifts up so we can add a second figure to show the effect of changing δ.  We first show the one-sided p-value which corresponds to ph when δ=0.  The initial value of ph=0.054 so the value forpb=1−ph=0.946.  The value of ph is shaded red because it is greater than the harm threshold ηh=0.05.   We then add ph whenδ=0; pb is much greater than the beneficial threshold ηb=0.25 so is shaded light blue (so we can distinguish it later).  The combined conclusion of red and blue is Uncertain (which was represented in purple).  

As δ increases, pb and ph decrease.  As ph decreases to below ηh=0.05, the shading of the area representing ph switches to blue.  With both values shaded blue, the conclusion is that there is a Beneficial effect.  As δ continues to increase, pb eventually decreases to less than ηb=0.25, at which point the shading becomes grey and the conclusion is that the effect is Trivial.  The figure shifts up and simultaneously reduces δ to return to the start of the previous sequence so we can add an additional figure.  This figure represents the posterior distribution of the difference in means.  The final sequence shows the effect of increasing δ on the posterior probabilities that the difference in means is less than −δ and greater than δ respectively.  Compared to the p-values, the areas are in the opposite tails and there is a single posterior distribution rather than two sampling distributions.

 

Research areas

Statistical Inference, Statistical Modelling, Robustness, Nonparametric and semiparametric methods, Analysis of Sample Surveys, Ecological Monitoring.

 

Updated:   1 March 2017 / Responsible Officer:  CBE Communications and Outreach / Page Contact:  College Web Team