We Have a Pulse

…but not much more than that. I am not dead, but I am snowed under by a variety of contractual and literary obligations, and if anyone out there really wants to free up enough of my time for more frequent postings here on the ol’ ‘crawl, they’ll show me an easy way to calculate the variance of a population estimate based on stratified strip transects of unequal length, when y along each transect has already been converted into a distance-weighted mean-density value prior to the variance calculation*.

But in the meantime, the galleys for Hillcrest V. Velikovsky just came in from Nature, and I really like the (unaccredited) illo by Jason Cook (thanks for the link, Henry) so I’m posting it here as a placeholder, along with a brief excerpt:

Mr Velikovsky was obviously well-versed in placebo effects, having built an erudite display on the subject. What did he think would happen, the Prosecution thundered, when he forced his so-called “truth” down the throat of someone whose motto — knitted into her favourite throw-cushion — was If ye have faith the size of a mustard seed, ye shall move mountains? In telling ‘the truth’ Velikovsky had knowingly and recklessly endangered the very life of another human being.

Velikovsky pointed out that he hadn’t even known Lacey Hillcrest existed, adding that needlepointing something onto a pillowcase did not necessarily make it true. The Prosecution responded that the man who plants land mines in a playground doesn’t know the names of his victims either, and asked if the defendant’s needle-point remark meant that he was now calling Jesus a liar. The Defense objected repeatedly throughout.

I initially wrote this piece as parody. Judging by some of the wacko responses to last month’s podcast over on Starship Sofa, however, maybe I should reconsider.

More later. When I have, you know, a life.

*I mean, seriously, what are you supposed to use for n? Number of transects surveyed? The count has already been converted into units-per-square-mile. Number of square miles surveyed? Then how are you supposed to quantify variance between square miles, when each transect covers many miles and there’s no way to position sightings within each transect?

And why are these bloody Americans still using “square miles” anyway? Next they’ll be telling me to express transect length in furlongs…

This entry was written by Peter Watts , posted on Friday July 04 2008at 08:07 am , filed under writing news . Bookmark the permalink . Post a comment below or leave a trackback: Trackback URL.

7 Responses to “We Have a Pulse”

  1. Sorry Peter. My knowledge of statistics peaked at the difference between a sample and a population and the difference between a mean and a median.

    Although I have done research on the effects of exercise and I have discovered that 100% of people who exercise, die. I have also been told that 100% of people that don’t exercise also die but I can’t see putting in all that effort when the outcome is the same.

  2. Obvious question, but can’t you get the raw data?

    – Lars

  3. Mr. Watts.

    I have a partial solution for you little problem:

    1 square miles = 2.58998811 square kilometers.

    YOU ARE WELCOME!

    As for the rest of that shit…eh, you’ll have to use your freaky canadian juju to sort that out.

  4. Hi Peter – our artist is called Jacey aka Jason Cook

    http://www.jasoncook.co.uk/

    and he’s been our regular artist since 2000.

  5. It LIVES!

    Pardon me, I am in a good mood today, so I must do this:

    😀

    Anyway – reviewing a paper for a buddy, maybe? I have to agree with Lars, I’d be tempted to ask my friend to let me peek at the actual data! Without seeing how the buddy did his/her calculating to date, it’s soooo confusing, and could lead to forehead-slapping error.

    I mean, seriously, what are you supposed to use for n? Number of transects surveyed?

    I’m picturing your buddy as looking at how some bio-thingy changes with depth, or how far from shore, so he sampled at various depths and distances.

    a. Are the transects considered single samples, and you have a bunch of ’em in the area you want to designate as Group 1, and another set of transects in Group 2, Group 3, etc.? Then I might call the number of transects “n”. Is he wanting to do an ANOVA? Does he have a certain test in mind?

    b. Or does each transect stand in for a bunch of estimated samples? So that the transect is iteslf Group 1. (Gonna depend on what you wanna check for differences between, I guess?)

    c. Also, important to pin down – if the transect is the slice between two actual samples, and he estimated between them, did he do a straight line kind of estimation, as in, the mid-point between them physically as the average of the two samples?
    Gawd, especially since the transects are different *lengths*.

    You must be some kinda friend, helping him/her untangle this stuff …

    HALP!! We need a Stats Prof! STAT!

    Good luck on this.

  6. Well, I have a better understanding of statistics than the average molecular or developmental biologist, but that is saying frighteningly little (if you can explain the difference between standard error and standard deviation beyond mentioning the fact that standard error bars are smaller, you can make the same claim), but I’m not sure what you mean by “a distance-weighted mean-density value”.

    If you mean that you have just one value for mean population density over each transect along with the size of each transect the simple stats version is to treat each transect as one sample.

    If you have a more advanced method which somehow uses the size of the transects but assumes them to all be the same size I think you could safely pretend they are all of the smallest size and sacrifice statistical power without introducing bias.

    If you have density values at points along the transects you could choose, at random, one such reading from each transect to generate a set of independent population estimates with n being the number of transects. That throws out even more power, but seems even safer. You are retroactively turning the transects into point samples at least as random as the original data.

    That is all assuming you really do just want the variance of the population estimate, and not some kind of measure of the actual variation in population density at some particular scale (sounds like miles).

    In that case I’m not sure it is possible if all you have is the average density of each transect. The variation at a particular scale calculation doesn’t seem possible with samples of different scales.

  7. I do in fact have access to the raw data, but the gummint department providing it has already pledged fealty to a certain formula which transforms the values into something else, scaled across different areas, and it’s these transformed values I’m trying to get the variance for. Also the formula itself may produce biased results, according to a 2000 paper I’ve just dug up.

    But life is good, I think. That same paper showed me a technique for generating the same estimate, but much better measures of variation around it, so I am happy. Or will be until the contractors come back to me and tell me they don’t like the results I’ve presented. So far it’s been over 24 hours since I submitted the revised analysis, and there’s been nothing but silence.

    I find that a bit ominous…