Input data quality control for NDNQI national comparative statistics and quarterly reports: a contrast of three robust scale estimators for multiple outlier detection

Table 3 False alarm rate as a function of skewness in data distribution for IQR, MAD, or FAST-MCD approach with simulation

Asymmetry in data distribution		Potential outlier rate by different methods
Preset skewness	Estimated skewness	Percentile	IQR	MAD	FAST-MCD
0.000	−0.01(0.077)	0.025(0.005)	0.026(0.007)	0.023(0.007)	0.027(0.008)
1.000	0.983(0.116)	0.025(0.005)	0.035(0.006)	0.036(0.006)	0.078(0.013)
1.414	1.398(0.174)	0.025(0.005)	0.045(0.006)	0.049(0.007)	0.129(0.015)
1.732	1.720(0.221)	0.025(0.005)	0.053(0.007)	0.062(0.007)	0.186(0.014)
2.000	1.959(0.249)	0.025(0.005)	0.060(0.007)	0.073(0.008)	0.227(0.014)
2.236	2.197(0.286)	0.025(0.005)	0.066(0.007)	0.085(0.008)	0.260(0.014)
2.449	2.397(0.318)	0.025(0.005)	0.072(0.007)	0.097(0.009)	0.288(0.014)
2.646	2.581(0.343)	0.025(0.005)	0.077(0.007)	0.108(0.009)	0.313(0.014)
2.828	2.759(0.375)	0.025(0.005)	0.082(0.007)	0.120(0.009)	0.333(0.013)
3.000	2.928(0.409)	0.025(0.005)	0.087(0.007)	0.132(0.009)	0.351(0.013)
3.162	3.076(0.433)	0.025(0.005)	0.092(0.007)	0.144(0.019)	0.367(0.013)
3.317	3.225(0.467)	0.025(0.005)	0.096(0.007)	0.194(0.101)	0.380(0.012)
3.464	3.364(0.502)	0.025(0.005)	0.099(0.007)	0.388(0.167)	0.392(0.011)

Mean rate of potential outliers with standard deviation in parenthesis for 1,000 simulated data sets at each preset skewness level.

ISSN: 1756-0500