Skip to main content

Table 1 Estimates of the parameters with mean squared errors (in parentheses) with the normality assumption for errors

From: Sample selection bias with multiple dependent selection rules: an application to survey data analysis with multilevel nonresponse

Case

Parameter

TRUE

BSM (two-step)

BSM (one-step)

USM (one-step)

CC

1

\(\alpha _{11}\)

4.5

4.54 (0.20)

4.54 (0.20)

1.95 (2.55)

4.54 (0.19)

\(\alpha _{12}\)

− 0.6

− 0.60 (0.03)

− 0.60 (0.03)

− 0.30 (0.31)

− 0.60 (0.03)

\(\alpha _{21}\)

1

1.05 (0.25)

1.05 (0.25)

–

0.69 (0.34)

\(\alpha _{22}\)

0

− 0.02 (0.07)

− 0.02 (0.07)

 

0.10 (0.10)

\(\beta _{1}\)

− 1

− 0.80 (0.22)

− 0.86 (0.23)

− 0.85 (0.19)

− 0.80 (0.23)

\(\beta _{2}\)

1.5

1.44 (0.07)

1.48 (0.05)

1.49 (0.06)

1.43 (0.07)

\(\rho _{01}\)

− 0.5

− 0.01 (0.39)

− 0.31 (0.28)

− 0.35 (0.37)

–

\(\rho _{02}\)

0

0.00 (0.00)

− 0.22 (0.38)

–

–

\(\rho _{12}\)

0

0.05 (0.34)

0.05 (0.37)

–

–

\(\sigma _{00}\)

1

0.96 (0.06)

1.05 (0.11)

1.10 (0.11)

0.97 (0.06)

2

\(\alpha _{11}\)

2

2.07 (0.17)

2.07 (0.17)

2.96 (0.98)

2.07 (0.17)

\(\alpha _{12}\)

− 0.2

− 0.21 (0.02)

− 0.21 (0.02)

− 0.45 (0.25)

− 0.21 (0.02)

\(\alpha _{21}\)

5

4.96 (0.72)

5.02 (0.64)

–

3.92 (1.11)

\(\alpha _{22}\)

− 0.7

− 0.69 (0.08)

− 0.69 (0.09)

–

− 0.43 (0.27)

\(\beta _1\)

− 1

− 0.96 (0.19)

− 1.01 (0.11)

− 1.04 (0.11)

− 0.83 (0.19)

\(\beta _2\)

1.5

1.48 (0.12)

1.49 (0.03)

1.52 (0.03)

1.41 (0.09)

\(\rho _{01}\)

− 0.5

− 0.32 (0.49)

− 0.26 (0.42)

− 0.60 (0.21)

–

\(\rho _{02}\)

− 0.5

− 0.10 (0.56)

− 0.47 (0.30)

–

–

\(\rho _{12}\)

0.5

0.63 (0.29)

0.47 (0.55)

–

–

\(\sigma _{00}\)

1

1.67 (2.73)

1.01 (0.10)

1.03 (0.09)

0.91 (0.11)

3

\(\alpha {11}\)

4.5

4.68 (0.45)

4.68 (0.45)

3.13 (2.61)

4.68 (0.45)

\(\alpha {12}\)

− 0.5

− 0.52 (0.05)

− 0.52 (0.05)

− 0.32 (0.33)

− 0.52 (0.05)

\(\alpha {21}\)

− 3

− 3.06 (0.26)

− 3.04 (0.27)

–

− 3.06 (0.26)

\(\alpha {22}\)

1

1.03 (0.09)

1.02 (0.09)

–

1.03 (0.09)

\(\beta {0}\)

− 1

− 0.65 (0.60)

− 1.00 (0.25)

− 0.38 (0.69)

− 0.20 (0.81)

\(\beta _{1}\)

1.5

1.43 (0.11)

1.50 (0.06)

1.37 (0.13)

1.36 (0.14)

\(\rho _{01}\)

− 0.5

0.12 (0.81)

− 0.42 (0.36)

0.63 (1.13)

–

\(\rho _{02}\)

0.5

0.12 (0.56)

0.63 (0.10)

–

–

\(\rho _{12}\)

− 0.5

− 0.50 (0.10)

− 0.51 (0.21)

–

–

\(\sigma _{00}\)

1

1.01 (0.17)

1.05 (0.08)

0.99 (0.14)

0.91 (0.10)

  1. Estimates are based on the bivariate selection model using two-step and one-step methods, BSM (two-step) and BSM (one-step), univariate selection model in one-step method, USM (one-step), and complete case analysis (CC), i.e., data after deleting nonresponse cases.
  2. The sample size is N = 1000.
  3. Case1: Random nonresponse at one level and MNAR at another level of nonresponse,
  4. Case2: MNAR in the same direction with the variable of interest at both levels of nonresponse,
  5. Case3: MNAR at both levels of nonresponse and with different direction with the variable of interest