Statistical odds of animal dieoffs

[BernieM]

I am sure probably everyone has seen the news about the flocks of birds and fish that are dying off at different places in the world. The point of my question here is not to somehow explain why they died or if it's the beginning of the apocalypse. Rather I am curious about whether there is a satisfactory way to calculate if the odds of such a thing happening are within statistical reason and probability.

I am sure that flocks of birds or schools of fish have died in the past plenty of times in the past hundreds of millions of years, especially in extinction events. Barring massive extinction events what is the probability that x number of die off events of different causes at different places on the globe will happen within y period of time?

i.e., 1 dieoff per year = 100% 2 dieoffs a year with different causes =98% ... 245 dieoffs a year with different causes =.0001%

 

[bpet]

One approach is to use a Poisson model, i.e. test the null hypothesis that dieoff events occur independently at a particular rate; of course the difficult part is getting enough data for a reasonable estimate of the rate of dieoff events.

 

[BernieM]

Yes, I realized after posting that one had to have some data. I guess that would present a big problem in calculating this.

 

[BernieM]

Found the US data here for the past year. As I suspected of course, the reason it appears to be the apocalypse is because the media is reporting heavily, what they didn't report on frequently in the past. So I guess that I can show that the media reporting of these events is statistically out of the ordinary but animal dieoffs aren't.

http://www.nwhc.usgs.gov/mortality_events/ongoing.jsp

Given this data however, at what number of events per time period WOULD they become statistically significant as an 'unusual' event. Is there really a mathematical black and white line? Or does it become more of a question of what someone wants to define as 'significant' ?

 

[SW VandeCarr]

Found the US data here for the past year.

Given this data however, at what number of events per time period WOULD they become statistically significant as an 'unusual' event. Is there really a mathematical black and white line? Or does it become more of a question of what someone wants to define as 'significant' ?

There are many statistical tests for trend and for time series analysis. The problem here is the data. Your data is only for one year. I would say you would need some 10 years to overcome any recent reporting bias. A good test is the Cochran-Armitage test for trend.

I would simply take the number of reports by year (not weighted in any way by severity since this would be affected by local conditions) and see if there is a statistically significant long term trend of reported animal die offs. If you think there's a reporting bias in 2010, you could leave out that year. This would be very superficial, but would indicate whether you are dealing with a long term trend. Short term, you can't really draw any conclusions regarding a spike in 2010 based on a statistical tend analysis.

BTW, the Mayan "Long Count" is based in part on a period of 240 days which is very specific to the latitude. That number will vary from the Tropic of Cancer to the equator and is related to whether the subsolar point is north or south of a particular location. Outside the tropics the sun is always in the southern half of the sky in the northern hemisphere. So it's hard to see how a cycle that is dependent on a specific location could have world wide significance.

EDIT: You could also of course do a chi square test to see if the number of reported incidents in government records in 2010 is significantly larger than in the previous 10 year expectation..

 

[EnumaElish]

I'm thinking along the lines of: (1) using multiple regression analysis (controlling for broad species, broad area, etc.) establish a baseline of "explained deaths" based on deaths per day (or per week, etc.) whose causes have been determined; (2) use the baseline model to predict the number of deaths per day (of a broad species in a broad area) whose causes are (a) undetermined, or (b) open, or (c) suspect, [separately for a, b, c or a + b + c together or in any other combination]; (3) test whether the difference actual minus predicted deaths per day is stat. significant. Each significant difference could be interpreted as an extraordinary (although not necessarily supernatural) event.

 

[SW VandeCarr]

I'm thinking along the lines of: (1) using multiple regression analysis (controlling for broad species, broad area, etc.) establish a baseline of "explained deaths" based on deaths per day (or per week, etc.) whose causes have been determined; (2) use the baseline model to predict the number of deaths per day (of a broad species in a broad area) whose causes are (a) undetermined, or (b) open, or (c) suspect, [separately for a, b, c or a + b + c together or in any other combination]; (3) test whether the difference actual minus predicted deaths per day is stat. significant. Each significant difference could be interpreted as an extraordinary (although not necessarily supernatural) event.

And the OP is going to do all of this? I suggested something the OP might be able to do if s/he has access to ten years of data similar to the 2010 data to see if there is there is broad evidence of a problem.

 

[BernieM]

I found data going back to 1995 listed in 'Quarterly Reports' on the USGS website.

http://www.nwhc.usgs.gov/publications/quarterly_reports/index.jsp

In addition I was able to glean some basic information stating that at any time in the USA (depending on time of year, i.e., migration) that there are between 10 and 20 billion birds, among which about 1/2 will die during the year. I am not sure if that is as helpful.

From the looks of it, at first glance, it seems more a case of an 'extraordinary reporting incident' than an extraordinary dieoff event.

 

[BernieM]

Ok, I have taken the data on the USGS Wildlife Health website and summed up by # of reported dieoff events per year and total # of dieoff events in that year => 100 animals. Here is the data.

YEAR, # of of dieoff incidents , # of reports =>100 animals that year
2010 146 58
2009 171 52
2008 194 65
2007 131 51
2006 148 62
2005 154 54
2004 131 45
2003 164 57
2002 161 53
2001 155 58
2000 171 70
1999 144 63
1998 141 70
1997 141 74
1996 146 59
1995 177 62

Well, I don't know about anyone else but these numbers aren't screaming Apocalypse to me!

 

[EnumaElish]

Do you have # of deaths?

 

[BernieM]

The last column is the total number of reports per year involving 100 or more deaths per incident. The total number of deaths per year I don't have but could get, but to be honest having looked at all the data already, there isn't really anything unusual in total number of deaths per year either. I will get that data however and post it here.

 

[EnumaElish]

I for one will greatly appreciate any further information you can post here; for example the geography and the seasonality of the deaths each year.

 

[BernieM]

You can get the additional information here:

http://www.nwhc.usgs.gov/publications/quarterly_reports/index.jsp

and here

http://www.nwhc.usgs.gov/mortality_events/ongoing.jsp

 

[EnumaElish]

Of course, the smarter thing to do for anyone who would invest resources into downloading and summarizing the data is to apply for a research grant.

 

[BernieM]

Additional info. Following is by year # of dieoff events =>1,000 animals, =>10,000, =>100,000 and =>1,000,000 animals:

2010 15 1 0 0
2009 18 5 3 1
2008 19 1 0 0
2007 21 1 0 0
2006 22 2 0 0
2005 22 0 0 0
2004 12 1 0 0
2003 18 2 0 0
2002 15 2 1 0
2001 14 3 0 0
2000 13 0 0 0
1999 21 3 0 0
1998 36 5 0 0
1997 29 7 2 0
1996 20 5 2 0
1995 18 4 1 0

 

[BernieM]

So now that I have some data, how would I show that current events are likely normal events not out of the ordinary. If I was Al Gore how would I show that the end of the world was coming? If I was a conspiracy theorist how would I show from the data that the data is all a big government coverup conspiracy?

 

[SW VandeCarr]

So now that I have some data, how would I show that current events are likely normal events not out of the ordinary. If I was Al Gore how would I show that the end of the world was coming? If I was a conspiracy theorist how would I show from the data that the data is all a big government coverup conspiracy?

You can't prove a negative with statistics. What you can say is that there is no evidence for a significant multi-year trend of animal die-offs in the overall numbers (across species) reported to the US government for the past 15 years. Those that take the other side will need to show data to back their argument. Theories about how well the reported data reflects reality is not a sufficient basis for an argument without data to back an alternative hypothesis.

 

[BernieM]

Well I posed the last challenge, to promote a little fun and see how creatively and entertainingly the facts could be statistically 'manipulated' so to speak.

 

[BernieM]

I graphed the data, I don't see anything significant in it. But maybe someone with a trained eye does?

Graph is in attachment.