I have a loglinear model of:

$$log(\mu(S_{ij|gij}))=\alpha_0+\alpha_1g_{ij}$$

where gij is distance, and Sij is connectivity

There is a bias in the distribution of count values for the outcome variable (Sij): meaning when values of gij (distance) are low then the Sij (connectivity) will be high (There is a bias for the connectivity relating to the distances, the shorter the distance the stronger the connectivity)

This BIAS only exists for distances shorter than 17mm. If distances were greater than 17mm the bias is negligible.

How can I adjust for this bias in my loglinear model?