(10)
Lj= α, + β Xj+ εj.
Finally, the random effects specification is similar to equation (9), but the disturbance term
includes an unobserved, random and state-specific effectμi :
(11) Lj = αo + β Xj + μ, + εj.
The dependent variable Lij is defined as the percentage overlap between i-th and j-th
states’ noxious weed (NXWS or NXW) regulations. The overlap data vector is of dimension
2304x1 (48x48 state-pair overlaps), which is constructed by transposing each row of overlap
matrix and stacking them into a column vector. Since diagonal elements of the percent overlap
matrix are equal to one, we delete i-th state’s overlap with its own list and consider observations
when i ≠ j. Thus, we have 2256 (48x47) observations on Lij for NXWS lists. Since NXW lists
apply to fewer states, there are only 870 and 1190 observations of state-pair overlaps in 1997 and
2002 NXW lists, respectively. We compute Lij for four lists: NXWS, NXWS prohibited, NXWS
restricted, and NXW lists. In the following, we refer to the overlap regression for each of the
above four lists as List 1 through 4, respectively.
Consistent with the previous section, the explanatory variables, Xij, fall into three groups:
(i) ecosystem dissimilarities between i-th and j-th state (Iij) in terms of average temperature and
precipitation, variance of temperature and precipitation, soil and land types, and water sources (7
variables), (ii) agronomic dissimilarities between i-th and j-th state (Aij) captured by field crops’
and irrigated area share of total crop land (2 variables), and (iii) lobbying dissimilarities between
i-th and j-th states (ωij) represented by contributions of seed producers, commodity producers
and the consumer groups (3 variables). Thus, the 1x12 vector of explanatory variables is given
by X = [ ιj,..., ιj, Aj, A2, ωc, ωis, ωj ].
16