Population aggregates from administrative data samples–how good are they?

Journal for Labour Market Research

Table 2 Quality indicators for cell deviations from target (full population version of BHP)—district#3-digit-industry level

	(i)	(ii)	(iii)	(iv)	(v)
	Workers			Establishments
	BHP 50%	SIAB Individual File	SIAB Basis Establishment File	BHP 50%	SIAB Basis Establishment File
Absolute deviation
Mean	− 2.5	− 1.0	− 1.1	0.0	− 0.4
p-value	0	0	0	0.26	0
rmse	596.3	148.8	101.3	6.1	31.7
mae	144.1	82.9	59.1	3.4	14.1
Ratio	0.323	0.186	0.132	0.092	0.376
Percentage deviation
Mean	0	0	0	0	− 0.001
p-value	0.679	0.893	0.907	0.438	0.793
mape	0.519	0.787	0.739	0.372	0.953
Percentage deviation (size weighted)
Mean	− 0.006	− 0.002	− 0.002	0	− 0.01
p-value	0	0	0	0.26	0
mape	0.323	0.186	0.132	0.092	0.376
N	789,616	789,616	789,616	789,616	789,616

The table shows various quality indicators for cell deviations from the target dataset (the full population version of the BHP) at the district#3-digit-industry level, including absolute deviations, percentage deviations and percentage deviations weighted by cell size. Approximations are calculated for the number of workers and establishments. Calculations are based on the 50 percent sample of the BHP, the SIAB Individual File and the SIAB Basis Establishment File, respectively. Indicators are the mean error (mean), root mean squared error (rmse), mean absolute error (mae), mean absolute percentage error (mape) and ratio of the total sum of errors to the total sum of cell counts (ratio). p-values for a t-test of (mean) against zero are shown in (p-value)
Sources: Establishment History Panel (BHP)—Version 7519 v2 (https://doi.org/10.5164/IAB.BHP7519.de.en.v2); Sample of Integrated Labour Market Biographies (SIAB)—Version 7519 v1 (https://doi.org/10.5164/IAB.SIAB7519.de.en.v1), own calculations