30.10.2012 Views

Proceedings of the Seventh Mountain Lion Workshop

Proceedings of the Seventh Mountain Lion Workshop

Proceedings of the Seventh Mountain Lion Workshop

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

except that attacks by pumas on humans are<br />

rare. We know nothing <strong>of</strong> bias caused by<br />

missing data. We can speculate that missing<br />

cases may not have been considered<br />

important enough to report; <strong>the</strong>re may have<br />

been nobody to report to; fatal incidents may<br />

have been undiscovered; <strong>the</strong>y may have<br />

been reported and <strong>the</strong> records lost; <strong>the</strong>y may<br />

have been printed in obscure references that<br />

we did not find; we may have wrongly<br />

discarded some incidents recorded from<br />

word-<strong>of</strong>-mouth accounts, etc. Thus, we<br />

have a core <strong>of</strong> strictly defined data that we<br />

treat as a total count. These data are<br />

restricted by <strong>the</strong> verification criterion in<br />

such a way that <strong>the</strong> passage <strong>of</strong> time reduces<br />

opportunity for verification. The core exists<br />

in a matrix <strong>of</strong> less well-defined incidents,<br />

<strong>the</strong> statistical properties <strong>of</strong> which are<br />

unknown.<br />

At a finer level, even <strong>the</strong> welldocumented<br />

cases have many blank cells in<br />

<strong>the</strong> data matrix because specific items were<br />

unknown or not reported, and <strong>the</strong>re was no<br />

way to infer <strong>the</strong> information. In <strong>the</strong>se<br />

situations, we usually cannot assume <strong>the</strong><br />

nature <strong>of</strong> possible biases caused by missing<br />

data, if <strong>the</strong>y exist.<br />

To summarize, we have nearly total<br />

counts <strong>of</strong> incidents in Beier-quality<br />

categories, missing incidents in o<strong>the</strong>rs, and<br />

missing data in all incidents. The categories<br />

in which we can assume total counts may be<br />

subject to a time-related bias that may affect<br />

<strong>the</strong> statistical distribution <strong>of</strong> <strong>the</strong> data. We<br />

have little a priori information to guide us.<br />

Therefore, we are restricted mostly to<br />

descriptive statistics and forming hypo<strong>the</strong>ses<br />

that may later be independently verified.<br />

We do use a Chi-square test to explore <strong>the</strong><br />

similarity <strong>of</strong> <strong>the</strong> distributions in three<br />

different categories, two <strong>of</strong> which we<br />

assume are total counts.<br />

Statistical Methods<br />

We have no a priori models or<br />

hypo<strong>the</strong>ses. We had believed that most, if<br />

PROCEEDINGS OF THE SEVENTH MOUNTAIN LION WORKSHOP<br />

REDUCING PUMA ATTACKS · Fitzhugh et al. 93<br />

not all, puma attacks were predatory, but<br />

information provided by Sweanor et al.<br />

(personal communication) contradicts that<br />

belief. Thus, our analysis is exploratory,<br />

examining <strong>the</strong> data to find hypo<strong>the</strong>ses and<br />

descriptive models that may later be<br />

subjected to data collection and statistical<br />

interpretation. In <strong>the</strong> few cases where<br />

statistical testing was warranted, we report<br />

<strong>the</strong> test used along with <strong>the</strong> results, but for<br />

<strong>the</strong> majority <strong>of</strong> situations, descriptive<br />

statistics are <strong>the</strong> only analysis used.<br />

Never<strong>the</strong>less, we feel secure in drawing<br />

some conclusions about how to reduce risk<br />

<strong>of</strong> serious injury during a puma encounter.<br />

RESULTS AND DISCUSSION<br />

What Can We Tell About <strong>the</strong> Data?<br />

With respect to counting attacks and<br />

comparing frequencies, we can be a little<br />

more specific about statistical qualities <strong>of</strong><br />

<strong>the</strong> data. Like Beier (1991), we believe we<br />

have a near-complete count <strong>of</strong> verified fatal<br />

attacks from 1890 to 2003 in <strong>the</strong> U. S. and<br />

Canada. Authors <strong>of</strong> new books (Danz,<br />

1999, Deurbrouck and Miller 2001, and<br />

Etling 2001) did extensive new searches,<br />

and failed to find any fatal attacks that meet<br />

Beier’s criteria that were not included by<br />

Beier in his original list, or else occurred<br />

after his publication. The 108 attacks we<br />

analyze that meet Beier’s criteria include 7<br />

fatal and 38 non-fatal attacks that occurred<br />

after Beier published his list, and 9 fatal and<br />

54 non-fatal attacks that meet Beier’s<br />

criteria and occurred between 1890 and<br />

1991. On <strong>the</strong> basis <strong>of</strong> additional information<br />

(personal communication, Dale Elliot to<br />

Etling, July 2000), we moved one <strong>of</strong> Beier’s<br />

non-fatal attacks (Bird and Sieh, Nevada,<br />

1971) to <strong>the</strong> “provoked attack” category and<br />

added 11 new non-fatal attacks between<br />

1890 and 1991. It is possible that attacks,<br />

even fatal ones, occurred in <strong>the</strong> U. S. that<br />

were never known, especially during <strong>the</strong><br />

depression years <strong>of</strong> <strong>the</strong> 1930s and <strong>the</strong><br />

various gold rushes in localities in <strong>the</strong>

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!