**Appendix A**

To simplify the analysis, but without loss of generality, we assume that we consider a supporting vector *b*<sup>β</sup> that contains *M* = 2 symmetric values, namely –*m* and *m*, with respective prior probabilities *q* β <sup>1</sup> and *q* β <sup>2</sup>. Consequently:

$$\mathbf{j}\_{ij}^{0} = \begin{bmatrix} -m\boldsymbol{q}\_1^{\beta} + m\boldsymbol{q}\_2^{\beta} \end{bmatrix} \mathbf{x}\_{ij} = \begin{bmatrix} -m\boldsymbol{q}\_1^{\beta} + m(\mathbf{1} - \boldsymbol{q}\_1^{\beta}) \end{bmatrix} \mathbf{x}\_{ij} \tag{A1}$$

The a priori probability distribution *q*<sup>β</sup> that guaranties that *y*ˆ<sup>0</sup> *ij* = *xij* is:

$$\mathbf{x}\_{ij}^{0} = \left[ -mq\_1^{\beta} + m\left(1 - q\_1^{\beta}\right) \right] \mathbf{x}\_{ij} \tag{A2}$$

$$1 = \left[ -mq\_1^{\beta} + m(1 - q\_1^{\beta}) \right] = q\_1^{\beta}(-2m) + m \tag{A3}$$

and the solution is:

$$\begin{array}{c} q\_1^\beta = \frac{m-1}{2m} \\ \text{and} \\ q\_2^\beta = 1 - q\_1^\beta = 1 - \frac{m-1}{2m} = \frac{m+1}{2m} \end{array} \tag{A4}$$

whereas in the GME-DWR approach, the prior used for these parameters is *q* β <sup>1</sup>= *q* β <sup>2</sup> <sup>=</sup> <sup>1</sup> 2 .

#### **Appendix B Analysis of the Sensitivity of the Estimates**

**Table A1.** DWP estimates on disaggregated mean annual wages (EUR) by industry, type of working day and gender, 2011. Support vectors as *b*-= [−10,0,10].


**Table A2.** DWP estimates on disaggregated mean annual wages (EUR) by industry, type of working day and gender, 2011. Support vectors as *b*-= [−1,000,0,1,000].

