Efficiencies of Inhomogeneity-Detection Algorithms: Comparison of Different Detection Methods and Efficiency Measures

Table 2

Properties of the test datasets.

	𝑓(CH)	𝑓(trend)	𝑓(PFa)	𝑓(PFb)	𝑓^′	𝑓	𝑠	𝑅
CH1B0	1.0	0	0	0	1.0	1.0	0	—
CH5B0	5.0	0	0	0	5.0	5.0	3.50	—
PF5B0	0	0	2.5	2.5	5.0	9.2	3.50	—
HUSTR	3.1	2.3	5.4	17.9	16.2	31.1	1.20	𝑅≥0.4
HUST0	2.9	2.4	5.1	17.9	15.5	30.3	1.02	—
CH5S0	5.0	0	0	0	5.0	5.0	1.00	—
PF5S0	0	0	2.5	2.5	5.0	9.2	1.00	—
CH5SR	5.8	0	0	0	5.8	5.8	1.15	𝑅≥0.4
PF5SR	0	0	3.9	2.7	7.8	12.2	1.34	𝑅≥0.4
CHPF0	3.3	1.5	1.4	1.4	7.6	10.4	0.0…4.0	—

𝑓(CH): frequency of change-points that are not parts of platforms, 𝑓(trend): frequency of trends, 𝑓(Pfa): frequency of platforms with longer than 5 years duration, 𝑓(Pfb): frequency of platforms with maximum 5 years duration, 𝑓^′: frequency of all but Pfb IHs, 𝑓: frequency of all IHs, s: standard deviation of IH-sizes, and 𝑅: restriction for the autocorrelation. Each frequency characteristic is shown as number per time series. Frequency characteristics show the frequency of introduced IHs during the generation, except for column 𝑓, where the empirical total frequencies are presented. (𝑓 is often slightly lower than the frequency of all introduced IHs, due to superposition of IHs or stretch-out over the end of the time series.)