Calculating Overall Relative Bias I am in some trouble to understand how is to calculate the overall relative bias. In this link, there are results of overall relative bias in  "Parameter estimates" sub-section under "Results" section. There they mentioned that :

the overall biases for the fixed effect parameters $\gamma_{00}$, $\gamma_{01}$, $\gamma_{10}$, and $\gamma_{11}$ were 0.6%, 2.6%, 1.4%, and 3.7% respectively. 

Could you please explain me how did they calculate it? I had thought that if I calculate the mean of the column of $\gamma_{00}$ in table 1, that is, $(8.77+4.75+3.94+\ldots-0.06-0.14)/27$ , I would get overall mean of $\gamma_{00}$. But it doesn't match with the result.
Also, they wrote that 

When the size of the group was increased to 30 with 30 groups, the bias was reduced to less than 6%. 

I am not understanding how have they  calculated that the  reduced bias is 6% in this case?
 A: The two questions here are about the numbers in Table 1 and the interpretation. I did some trivial investigations, but can only explain the second question.
For the second question,

When the size of the group was increased to 30 with 30 groups, the bias was reduced to less than 6%.

When Number of groups = 30, Group size = 30, let's pick up the four columns of  relative bias for fixed-effects estimates, as shown below. The maximum absolute value is 5.74%. So that's why "the bias was reduced to less than 6%".
γ00     γ01     γ10     γ11
0.07    1.09    -1.93   3.57
-0.08   3.70    -1.60   3.44
-0.18   5.74    -1.72   5.31

For the first question,

the overall biases for the fixed effect parameters γ00, γ01, γ10, and
  γ11 were 0.6%, 2.6%, 1.4%, and 3.7% respectively.

you're right, the average relative biases obtained from Table 1 are 
0.9 3.0 1.8 4.2.

and the differences with the claimed ones are
0.3 0.4 0.4 0.5.

The average differences with respect to the 27 numbers are
0.00957476  0.015185185 0.014910837 0.016995885.

My first guess is the rounding error. But the average rounding error should not be greater than 0.005. So it's possible that the estimates are rounded first, and the rounding error is amplified by the true values, which are equal or less than 1 in the absolute scale. The authors noted that "data not shown" after this quoted sentence, but it's unclear what data is not shown.
