This research is not conclusive in all respects, but it can certainly offer some implications so far as land-use modeling in northern Minnesota is concerned.  First, when examining the regressions for the year 1970 and the year 1990, it seems apparent that not all the influences on what percentage of land in a county is covered by forest were captured by this data.  Likewise, there appears to have been

a significant change in what factors influenced the percentage of forested land in a county at some point over the time period (1970-1990) in question.  However, both of these regression indicate an inverse relationship of one kind or another between agriculture and forestry, suggesting that a county where agriculture is more profitable (1970) or simply more common (1990) will have less in terms of amount of forested land.  A stronger R-squared value (the amount of variation in the independent variable being explained) is desirable, of course, to which end, I recommend that future studies perhaps attempt to include other variables that may have an effect on the amount of forested land in a county at any one time period, suggestions here would include variables concerning topography (such as the percentage of land at a steep grade), climate (average rainfall), or soil quality.

One procedural caution here must be addressed.  As one might expect, the percentage of a county devoted to agriculture is highly correlated with the profitability of agriculture in the county, which raises the concern of multicollinearity, where two independent variables are not independent and therefore capture much of the same variation in the dependent variable.  In cases of multicollinearity, however, the proper approach is to throw out whichever one of the variables is less significant in the model.  In both models, this was done, because one of the variables did not ever appear significantly correlated with the percent of forested land variable, thus neither restricted model should be at risk of multicollinearity. 

Both the regressions modeling the change in forested land cover from 1970 to 1990 explain roughly the same amount of variation in the independent variable, at around 55%.  However, the level-based model indicates that changes in population density and amount of pastured land are significantly correlated with the forested land change, while the percentage based model indicates that the percentage of forested land that remains in 1990 is only significantly correlated with the percent change in agricultural lands.  This difference illustrates an important caution where this kind of research is concerned.  The level-based models results seem to indicate that if one was interested in long term changes in the amount of forested land in northern MN, one should be more concerned with policy implications relating to population density or pastured lands than implications on how much of the land is being cultivated agriculturally.  This conclusion seems to run counter what our intuition might tell us, and additionally seem to run counter what both the regression based on only 1970 data and the regression based only on 1990 data suggest.  This is because using data series measured in levels for regressions of this nature is perhaps not very desirable; the largest counties would be expected to have the largest amount of forested land (and therefore, perhaps the largest changes in forested land), regardless what factors were influencing the percentage of land that was being forested.  Thus, the conclusions of this level-based change regression are likely muddled by the fact that county size is not adequately accounted for by the model.  The percentage-based change regression is considerably more desirable for this reason, as it examines the relative change in forestry based on the starting amount of forested land in a county.  Not surprisingly, this regression again confirms that changes in forested land are correlated with changes in agricultural land. Interestingly, however, the coefficient on the remaining agricultural lands variable is positive.  This seems to suggest that while in any given time period greater amounts of land devoted to agriculture are associated with smaller amounts of land covered by forests, in the long-run, over time, greater losses in agricultural lands are associated with greater losses of forest covered lands. 

These conclusions have significant implications for future forest conservation.  For those concerned simply with the amount of forested land existing in northern Minnesota, it appears that population density changes, changes in the percentage of land devoted to pasture, and even changes in the profitability of agriculture have little to no effect on changes in forested land.  Rather, it appears that the relative amount of land being cultivated agriculturally, compared to what had been cultivated before, is significantly correlated with changes in the amount of forested land in a county over time.  This might indicate, policy-wise, that programs such as farm subsidies and agricultural price supports could have a sort of third party effect on the amount of forested land in a county.  While long term changes in forestry are not correlated with changes in agricultural profitability, they are correlated with the changes in agricultural lands, which in turn are likely affected by the farm subsidies and price-supports.  Interestingly, the relationship, though, is not necessarily what one might expect; the higher the percentage of agricultural lands that are maintained over the time period, the larger the percentage of forest lands that are maintained over the same period.  This suggests if a policy goal is to augment forested land in the short term, the policy might need to address a trade-off between forests and farms.  But if the policy is intended to be long term, that trade-off disappears and instead the goal ought to be to preserve both the agricultural lands and the forests.

Of course, it is important to remember that a correlation does not imply that changes in the amount of land being farmed cause changes in the amount of land covered by forest.  However, it does suggest the two are related; policies interested in maintaining or increasing forest levels in the northwoods might therefore be advised to additionally concern themselves with the long and short term levels of agricultural cultivation in those same areas.