Reduced chi-squared statistic

In statistics, the reduced chi-square statistic is used extensively in goodness of fit testing. It is also known as mean squared weighted deviation (MSWD) in isotopic dating^[1] and variance of unit weight in the context of weighted least squares.^[2][3]

Its square root is called regression standard error,^[4] standard error of the regression,^[5][6] or standard error of the equation^[7] (see Ordinary least squares § Reduced chi-squared)

Definition

It is defined as chi-square per degree of freedom:^{[8][9][10][11]: 85 [12][13][14][15]} $${\displaystyle \chi _{\nu }^{2}={\frac {\chi ^{2}}{\nu }},}$$ where the chi-squared is a weighted sum of squared deviations: $${\displaystyle \chi ^{2}=\sum _{i}{\frac {(O_{i}-C_{i})^{2}}{\sigma _{i}^{2}}}}$$ with inputs: variance ${\displaystyle \sigma _{i}^{2}}$, observations O, and calculated data C.^[8] The degree of freedom, ${\displaystyle \nu =n-m}$, equals the number of observations n minus the number of fitted parameters m.

In weighted least squares, the definition is often written in matrix notation as $${\displaystyle \chi _{\nu }^{2}={\frac {r^{\mathrm {T} }Wr}{\nu }},}$$ where r is the vector of residuals, and W is the weight matrix, the inverse of the input (diagonal) covariance matrix of observations. If W is non-diagonal, then generalized least squares applies.

In ordinary least squares, the definition simplifies to: $${\displaystyle \chi _{\nu }^{2}={\frac {\mathrm {RSS} }{\nu }},}$$ $${\displaystyle \mathrm {RSS} =\sum r^{2},}$$ where the numerator is the residual sum of squares (RSS).

When the fit is just an ordinary mean, then ${\displaystyle \chi _{\nu }^{2}}$ equals the sample standard deviation.

Discussion

As a general rule, when the variance of the measurement error is known a priori, a ${\displaystyle \chi _{\nu }^{2}\gg 1}$ indicates a poor model fit. A ${\displaystyle \chi _{\nu }^{2}>1}$ indicates that the fit has not fully captured the data (or that the error variance has been underestimated). In principle, a value of ${\displaystyle \chi _{\nu }^{2}}$ around ${\displaystyle 1}$ indicates that the extent of the match between observations and estimates is in accord with the error variance. A ${\displaystyle \chi _{\nu }^{2}<1}$ indicates that the model is "overfitting" the data: either the model is improperly fitting noise, or the error variance has been overestimated.^[11]: 89

When the variance of the measurement error is only partially known, the reduced chi-squared may serve as a correction estimated a posteriori.

Applications

Geochronology

In geochronology, the MSWD is a measure of goodness of fit that takes into account the relative importance of both the internal and external reproducibility, with most common usage in isotopic dating.^{[16][17][1][18][19][20]}

In general when:

MSWD = 1 if the age data fit a univariate normal distribution in t (for the arithmetic mean age) or log(t) (for the geometric mean age) space, or if the compositional data fit a bivariate normal distribution in [log(U/He),log(Th/He)]-space (for the central age).

MSWD < 1 if the observed scatter is less than that predicted by the analytical uncertainties. In this case, the data are said to be "underdispersed", indicating that the analytical uncertainties were overestimated.

MSWD > 1 if the observed scatter exceeds that predicted by the analytical uncertainties. In this case, the data are said to be "overdispersed". This situation is the rule rather than the exception in (U-Th)/He geochronology, indicating an incomplete understanding of the isotope system. Several reasons have been proposed to explain the overdispersion of (U-Th)/He data, including unevenly distributed U-Th distributions and radiation damage.

Often the geochronologist will determine a series of age measurements on a single sample, with the measured value ${\displaystyle x_{i}}$ having a weighting ${\displaystyle w_{i}}$ and an associated error ${\displaystyle \sigma _{x_{i}}}$ for each age determination. As regards weighting, one can either weight all of the measured ages equally, or weight them by the proportion of the sample that they represent. For example, if two thirds of the sample was used for the first measurement and one third for the second and final measurement, then one might weight the first measurement twice that of the second.

The arithmetic mean of the age determinations is $${\displaystyle {\overline {x}}={\frac {\sum _{i=1}^{N}x_{i}}{N}},}$$ but this value can be misleading, unless each determination of the age is of equal significance.

When each measured value can be assumed to have the same weighting, or significance, the biased and unbiased (or "sample" and "population" respectively) estimators of the variance are computed as follows: $${\displaystyle \sigma ^{2}={\frac {\sum _{i=1}^{N}(x_{i}-{\overline {x}})^{2}}{N}}{\text{ and }}s^{2}={\frac {N}{N-1}}\cdot \sigma ^{2}={\frac {1}{N-1}}\cdot \sum _{i=1}^{N}(x_{i}-{\overline {x}})^{2}.}$$

The standard deviation is the square root of the variance.

When individual determinations of an age are not of equal significance, it is better to use a weighted mean to obtain an "average" age, as follows: $${\displaystyle {\overline {x}}^{*}={\frac {\sum _{i=1}^{N}w_{i}x_{i}}{\sum _{i=1}^{N}w_{i}}}.}$$

The biased weighted estimator of variance can be shown to be $${\displaystyle \sigma ^{2}={\frac {\sum _{i=1}^{N}w_{i}(x_{i}-{\overline {x}}^{*})^{2}}{\sum _{i=1}^{N}w_{i}}},}$$ which can be computed as $${\displaystyle \sigma ^{2}={\frac {\sum _{i=1}^{N}w_{i}x_{i}^{2}\cdot \sum _{i=1}^{N}w_{i}-{\big (}\sum _{i=1}^{N}w_{i}x_{i}{\big )}^{2}}{{\big (}\sum _{i=1}^{N}w_{i}{\big )}^{2}}}.}$$

The unbiased weighted estimator of the sample variance can be computed as follows: $${\displaystyle s^{2}={\frac {\sum _{i=1}^{N}w_{i}}{{\big (}\sum _{i=1}^{N}w_{i}{\big )}^{2}-\sum _{i=1}^{N}w_{i}^{2}}}\cdot {\sum _{i=1}^{N}w_{i}(x_{i}-{\overline {x}}^{*})^{2}}.}$$ Again, the corresponding standard deviation is the square root of the variance.

The unbiased weighted estimator of the sample variance can also be computed on the fly as follows: $${\displaystyle s^{2}={\frac {\sum _{i=1}^{N}w_{i}x_{i}^{2}\cdot \sum _{i=1}^{N}w_{i}-{\big (}\sum _{i=1}^{N}w_{i}x_{i}{\big )}^{2}}{{\big (}\sum _{i=1}^{N}w_{i}{\big )}^{2}-\sum _{i=1}^{N}w_{i}^{2}}}.}$$

The unweighted mean square of the weighted deviations (unweighted MSWD) can then be computed, as follows: $${\displaystyle {\text{MSWD}}_{u}={\frac {1}{N-1}}\cdot \sum _{i=1}^{N}{\frac {(x_{i}-{\overline {x}})^{2}}{\sigma _{x_{i}}^{2}}}.}$$

By analogy, the weighted mean square of the weighted deviations (weighted MSWD) can be computed as follows: $${\displaystyle {\text{MSWD}}_{w}={\frac {\sum _{i=1}^{N}w_{i}}{{\big (}\sum _{i=1}^{N}w_{i}{\big )}^{2}-\sum _{i=1}^{N}w_{i}^{2}}}\cdot \sum _{i=1}^{N}{\frac {w_{i}(x_{i}-{\overline {x}}^{*})^{2}}{(\sigma _{x_{i}})^{2}}}.}$$

Rasch analysis

In data analysis based on the Rasch model, the reduced chi-squared statistic is called the outfit mean-square statistic, and the information-weighted reduced chi-squared statistic is called the infit mean-square statistic.^[21]

References

1. Wendt, I., and Carl, C., 1991, The statistical distribution of the mean squared weighted deviation, Chemical Geology, 275–285.
2. Strang, Gilbert; Borre, Kae (1997). Linear algebra, geodesy, and GPS. Wellesley-Cambridge Press. p. 301. ISBN 9780961408862.
3. Koch, Karl-Rudolf (2013). Parameter Estimation and Hypothesis Testing in Linear Models. Springer Berlin Heidelberg. Section 3.2.5. ISBN 9783662039762.
4. Julian Faraway (2000), Practical Regression and Anova using R
5. Kenney, J.; Keeping, E. S. (1963). Mathematics of Statistics. van Nostrand. p. 187.
6. Zwillinger, D. (1995). Standard Mathematical Tables and Formulae. Chapman&Hall/CRC. p. 626. ISBN 0-8493-2479-3.
7. Hayashi, Fumio (2000). Econometrics. Princeton University Press. ISBN 0-691-01018-8.
8. Laub, Charlie; Kuhl, Tonya L. (n.d.), How Bad is Good? A Critical Look at the Fitting of Reflectivity Models using the Reduced Chi-Square Statistic (PDF), University California, Davis, archived from the original (PDF) on 6 October 2016, retrieved 30 May 2015
9. Taylor, John Robert (1997), An introduction to error analysis, University Science Books, p. 268
10. Kirkman, T. W. (n.d.), Chi-Square Curve Fitting, retrieved 30 May 2015
11. Bevington, Philip R. (1969), Data Reduction and Error Analysis for the Physical Sciences, New York: McGraw-Hill
12. Measurements and Their Uncertainties: A Practical Guide to Modern Error Analysis, By Ifan Hughes, Thomas Hase
13. Dealing with Uncertainties: A Guide to Error Analysis, By Manfred Drosg
14. Practical Statistics for Astronomers, By J. V. Wall, C. R. Jenkins
15. Computational Methods in Physics and Engineering, By Samuel Shaw Ming Wong
16. Dickin, A. P. 1995. Radiogenic Isotope Geology. Cambridge University Press, Cambridge, UK, 1995, ISBN 0-521-43151-4, ISBN 0-521-59891-5
17. McDougall, I. and Harrison, T. M. 1988. Geochronology and Thermochronology by the ⁴⁰Ar/³⁹Ar Method. Oxford University Press.
18. Lance P. Black, Sandra L. Kamo, Charlotte M. Allen, John N. Aleinikoff, Donald W. Davis, Russell J. Korsch, Chris Foudoulis 2003. TEMORA 1: a new zircon standard for Phanerozoic U–Pb geochronology. Chemical Geology 200, 155–170.
19. M. J. Streule, R. J. Phillips, M. P. Searle, D. J. Waters and M. S. A. Horstwood 2009. Evolution and chronology of the Pangong Metamorphic Complex adjacent to themodelling and U-Pb geochronology Karakoram Fault, Ladakh: constraints from thermobarometry, metamorphic modelling and U-Pb geochronology. Journal of the Geological Society 166, 919–932 doi:10.1144/0016-76492008-117
20. Roger Powell, Janet Hergt, Jon Woodhead 2002. Improving isochron calculations with robust statistics and the bootstrap. Chemical Geology 185, 191–204.
21. Linacre, J. M. (2002). "What do Infit and Outfit, Mean-square and Standardized mean?". Rasch Measurement Transactions. 16 (2): 878.