Statistics Glossary → Autocorrelation correction and prewhitening (Hamed-Rao, TFPW)
Autocorrelation correction and prewhitening (Hamed-Rao, TFPW)concept
Methods used with the Mann-Kendall trend test that stop positive autocorrelation in a series from misleading the test, either by correcting the variance of the test statistic or by removing the autocorrelation from the series.
The Mann-Kendall test assumes independent observations. In a positively autocorrelated series successive values resemble each other, so random fluctuations look like a trend; the variance of the test statistic S is underestimated and the chance of finding a significant trend where none exists rises. The Hamed and Rao (1998) method computes an 'effective sample size' ratio from the significant autocorrelations of the ranks of the detrended series and inflates Var(S) by that ratio. The Yue and Wang (2004) method makes a similar correction based on the lag-1 autocorrelation.
Prewhitening instead removes the autocorrelation from the series: the AR(1) component is filtered out with the transformation y_t - r₁·y_(t-1), and the test is applied to the 'whitened' series. When a trend is present, however, r₁ is overestimated and simple prewhitening removes part of the trend as well. Trend-free prewhitening (TFPW), proposed by Yue and colleagues (2002), first removes the trend using Sen's slope, estimates r₁ from the detrended series, filters out the AR(1) component and then adds the trend back. Which method to choose depends on the length of the series and its autocorrelation structure; reporting the methods side by side shows how sensitive the results are.
An everyday example
In 30 years of water-temperature measurements from a lake, a warm year is usually followed by another warm year, because the lake loses heat slowly. This 'memory' can make a few successive warm years look like a genuine warming trend. Autocorrelation correction accounts for this misleading effect of similar successive years and assesses more honestly whether the trend could be explained by chance.
In MerQur
It is chosen with the 'Autocorrelation correction (None (classical MK) / Hamed-Rao / Yue-Wang / TFPW (pre-whitening))' parameter in Mann-Kendall + Sen's Slope.
Analyses where this concept appears
Related concepts
Sen's slopeAutocorrelation (ACF) and partial autocorrelation (PACF) functions