Facebook
X (Twitter)
Reddit
Pinterest
Tumblr
LinkedInHawks v. humans: Making anxious circles in the sky
Article reviewed: Restoring forest raptors: Influence on human disturbance and forest condition on Northern Goshawks
By M. Morrison, R. Young, J. Romsos, and R. Golightly. Published in Restoration Ecology, Vol. 19 No. 2 pp. 273-279, accessible here.
The plot line: Suspecting that human interactions with goshawks are causing their decline (like has been observed in Europe), the researchers measured human activity (hiking) and development (roads and houses) differences between areas that are occupied frequently versus infrequently to see if there was a difference. Human activity was higher in areas not used as much by goshawks, but the difference was not detectable with statistical tests and two particularly popular hiking spots appeared to be the reason for the difference (although the authors don’t mention this). Road density did appear to be greater in areas that were not used as much by goshawks, but there were also differences in forest structure and elevation between frequently and infrequently occupied sites. Despite largely inconclusive results, management recommendations are made.
Relevant quote: “There is little reason… to restore structural conditions for the goshawk if human disturbance will negate any positive benefits.”
Relevance to landowners and stakeholders:
As I have observed when listening to landowners who manage their forest for wildlife, they often do so because they enjoy seeing wildlife (who doesn’t?). Seeing wildlife on land managed for wildlife is tremendously gratifying. There is a paradox, however, if the human love for wildlife is unrequited. Such may be the case with goshawks. Although this study did not find evidence to support that human interaction is correlated with goshawk frequency, it would not be surprising if there was such a relationship because of possible adverse physiological responses to human presence (especially if the humans are directly harassing them). So the primary relevance is… don’t harass goshawks when you are hiking!
Relevance to managers:
The relevance with respect to managing forests is limited because of the experimental design, low statistical power, and habitat differences between study sites. The study did suggest what seems to be an efficient design for monitoring goshawks. They found that an “occupancy index” (# of observations over time / # of surveys; frequency, in other words) was related to reproductive success. One could therefore measure occupancy index and assume that it is pretty well correlated with reproductive success. They also point out that it is important to survey multiple years because of wide year-to-year variability in breeding success.
Critique and/or limitation (there's always something, no matter how good the article is):
In the abstract and many times in the results and discussion, they state that human interaction was much higher in areas that were infrequently occupied by goshawks. Technically, this was true. But they also relied on a statistical test to inform them on whether or not they would get the same result consistently if they were to do the experiment again (i.e to tell them if the difference was “real”). The difference was far from what most people (even wildlife biologists who tend to have higher p value thresholds) would consider significant in a statistical sense (p=0.322). P-value thresholds (which they never defined) are subjective and in many cases useless, but if you use hypothesis testing, you can’t just dismiss non-significant results in some cases and accept them in others. The authors even make very broad management recommendations based on this non-significant result.
I wouldn’t critique this inference if the data actually did suggest that there was a real difference, but it doesn’t. Eight sites were used in the pool of frequently occupied territories and 13 in the pool of infrequently occupied territories. The reason they found a large difference in human activity (and the reason why the difference was not significant) was that two of the infrequently occupied sites were exceptionally popular hiking areas. This pushed the variance (and the p value) way up. We don’t know if these are true outliers or not because the sample size was small. But if you take these two sites out, there is virtually no difference in human activity between frequently and infrequently occupied territories. To make broad management recommendations from what appear to be inconclusive results seems far from appropriate.
Here are their data, with human interaction regressed against occupancy:
See those two points way up at the top? Those were exceptionally popular hiking areas with a lot of human interaction. When you take those out there is no relationship at all. Actually, when you leave them in there is still no relationship if your p-threshold (alpha) is 0.05 or even 0.10. And if you do accept it as a real relationship, the adjusted r2 value is a miniscule 0.07. Is that a close enough relationship to base a management recommendation on?
The counter-argument could be that when you consider all of the results, in sum, they add up to being enough to make a management recommendation. But when considering all the results, it just makes me think that there is even less reason for making management recommendations. The infrequently and frequently occupied sites had differences in forest structure. How do we know these differences are not the reason for the differences in occupancy frequency? They were not big differences in forest structure, but neither were the differences in human activity.
I do grant that there was a solid difference in road density between frequently and infrequently occupied areas. But there is no “weight of evidence” from many different results that suggests human activity in general has caused a decline in goshawk occupancy.
In the methods, they say that occupancy is not used as a continuous variable because of gap in the data, but then it is indeed used as a continuous variable to correlate occupancy with human activity, local road extent, and road+trail extent. How come the gap is not an issue with those variables? (you can kind of see the gap in the graph above between 0.5 and 0.8 on the x-axis).
They lead off the restoration implications with this: “Our results suggest that goshawk protection within the Basin has been insufficient and that some of the territories require actions to restore vegetation to pre-settlement conditions.” This is problematic: 1) They did not evaluate the sufficiency of protection measures. No overall measure of goshawk trend or status is made. The protection measures between occupied and unoccupied territories (their primary variable of interest) were the same. 2) They did not study how vegetation change since settlement has impacted goshawks. This is the first mention of pre-settlement conditions in the whole article.
Article reviewed: Restoring forest raptors: Influence on human disturbance and forest condition on Northern Goshawks
By M. Morrison, R. Young, J. Romsos, and R. Golightly. Published in Restoration Ecology, Vol. 19 No. 2 pp. 273-279
The plot line: Suspecting that human interactions with goshawks are causing their decline (like has been observed in Europe), the researchers measured human activity (hiking) and development (roads and houses) differences between areas that are occupied frequently versus infrequently to see if there was a difference. Human activity was higher in areas not used as much by goshawks, but the difference was not detectable with a statistical test and two particularly popular hiking spots appeared to be the reason for the difference (although the authors don’t mention this). Road density did appear to be greater in areas that were not used as much by goshawks, but there were also differences in forest structure and elevation between frequently and infrequently occupied sites. Despite largely inconclusive results, management recommendations are made.
Relevant quote: “There is little reason… to restore structural conditions for the goshawk if human disturbance will negate any positive benefits.”
Relevance to landowners and stakeholders:
As I have observed when listening to landowners who manage their forest for wildlife, they often do so because they enjoy seeing wildlife (who doesn’t?). Seeing wildlife on land managed for wildlife is tremendously gratifying. There is a paradox, however, if the human love for wildlife is unrequited. Such may be the case with goshawks. Although this study did not find evidence to support that human interaction is correlated with goshawk frequency, it would not be surprising if there was such a relationship because of possible adverse physiological responses to human presence (especially if the humans are directly harassing them). So the primary relevance is… don’t harass goshawks when you are hiking!
Relevance to managers:
The relevance with respect to managing forests is limited because of the experimental design, low statistical power, and habitat differences between study sites. The study did suggest what seems to be an efficient design for monitoring goshawks. They found that an “occupancy index” (# of observations over time / # of surveys; frequency, in other words) was related to reproductive success. One could therefore measure occupancy index and assume that it is pretty well correlated with reproductive success. They also point out that it is important to survey multiple years because of wide year-to-year variability in breeding success.
Critique and/or limitation (there's always something, no matter how good the article is):
In the abstract and many times in the results and discussion, they state that human interaction was much higher in areas that were infrequently occupied by goshawks. Technically, this was true. But they also relied on a statistical test to inform them on whether or not they would get the same result consistently if they were to do the experiment again (i.e to tell them if the difference was “real”). The difference was far from what most people (even wildlife biologists who tend to have higher p value thresholds) would consider significant in a statistical sense (p=0.322). P-value thresholds (which they never defined) are subjective and in many cases useless, but if you use hypothesis testing, you can’t just dismiss non-significant results in some cases and accept them in others. The authors even make very broad management recommendations based on this non-significant result.
I wouldn’t critique this inference if the data actually did suggest that there was a real difference, but it doesn’t. Eight sites were used in the pool of frequently occupied territories and 13 in the pool of infrequently occupied territories. The reason they found a large difference in human activity (and the reason why the difference was not significant) was that two of the infrequently occupied sites were exceptionally popular hiking areas. This pushed the variance (and the p value) way up. We don’t know if these are true outliers or not because the sample size was small. But if you take these two sites out, there is virtually no difference in human activity between frequently and infrequently occupied territories. To make broad management recommendations from what appear to be inconclusive results seems far from appropriate.
Here are their data, with human interaction regressed against occupancy:
See those two points way up at the top? Those were exceptionally popular hiking areas with a lot of human interaction. When you take those out there is no relationship at all. Actually, when you leave them in there is still no relationship if your p-threshold (alpha) is 0.05 or even 0.10. And if you do accept it as a real relationship, the adjusted r2 value is a miniscule 0.07. Is that a close enough relationship to base a management recommendation on?
The counter-argument could be that when you consider all of the results, in sum, they add up to being enough to make a management recommendation. But when considering all the results, it just makes me think that there is even less reason for making management recommendations. The infrequently and frequently occupied sites had differences in forest structure. How do we know these differences are not the reason for the differences in occupancy frequency? They were not big differences in forest structure, but neither were the differences in human activity.
I do grant that there was a solid difference in road density between frequently and infrequently occupied areas. But there is no “weight of evidence” from many different results that suggests human activity in general has caused a decline in goshawk occupancy.
In the methods, they say that occupancy is not used as a continuous variable because of gap in the data, but then it is indeed used as a continuous variable to correlate occupancy with human activity, local road extent, and road+trail extent. How come the gap is not an issue with those variables? (you can kind of see the gap in the graph above between 0.5 and 0.8 on the x-axis).
They lead off the restoration implications with this: “Our results suggest that goshawk protection within the Basin has been insufficient and that some of the territories require actions to restore vegetation to pre-settlement conditions.” This is problematic: 1) They did not evaluate protection measures. The protection measures between occupied and unoccupied territories (their primary variable of interest) were the same. 2) They did not study how vegetation change since settlement has impacted goshawks. This is the first mention of pre-settlement conditions in the whole article.
