understanding statistics

i dunno about you, but i barely do. (*^_^*)

if you’re like me, and you want to understand statistics better, have a look at emil kirkegaard’s understanding statistics page!

it’s a work in progress, but what he’s got there so far is really cool!

email says he’s got one planned for regression to the mean. any suggestions for other concepts?
_____

(p.s. – i am slowly working on my response to kevin macdonald. will try to get that up over the weekend, fingers crossed. buiyt nort whoile i’n tyyypoing. (~_^) )

(note: comments do not require an email. correlation≠causation)

14 Comments

  1. HBD Chick, Kevin Macdonald’s comments were so long that if i were you i would respond in a series of brief posts, rather than trying to respond to the whole thing at once:

    HBD chick responds to Kevin Macdonald part 1

    HBD chick responds to Kevin Macdonald part 2

    HBD chick responds to Kevin Macdonald part 3

    HBD chick responds to Kevin Macdonald part 4

    And you could discuss other topics in between these posts for variety

    Reply

  2. I have no idea who Kevin MacDonald is, what he wrote, or why you would need to respond to him. I have found that it is simply better to ignore most people on the internet, especially leftists. I only reply to people who seem to be genuinely interested in the truth, or for the edification of the audience. My guess is that whatever you write, it will only be read by people who read your blog. Most of us have no idea who this Kevin fellow is.

    Reply

  3. “Och, tell him you’re a Campbell and he can just piss off.”

    A few years ago I sent DNA into 23andme. As well as the analysis that 23andme gives you they also give you the raw data so you can send it to other places to be analyzed. I sent it to several places as well as to a professor named McDonald (an American) who kindly analyzed data for free. Around the time I sent mine I think he got a bit swamped . He did run my data through his program and sent back results saying it was 100% English, but also sent out emails telling people he was now more interested in getting more exotic DNA from native South Americans or something and asking people of strictly European decent to stop sending in DNA.

    I did send him one more email asking if by English he meant to include Scotland. He wrote back that he meant to include England , Scotland, Wales and Ireland. He just lumped them all in and called them English.

    Somewhere in the Highlands his McDonald ancestors are spinning in their graves. :-)

    Reply

  4. I have found that it is simply better to ignore most people on the internet, especially leftists.

    I wouldn’t necessarily consider Kevin MacDonald a leftist, unless you consider whites an oppressed group.

    Reply

  5. I think it would be an interesting visualization to show how multivariate regression (or correlation) influences bivariate relationships. That is, suppose X and Y are tightly correlated. But once you do a multivariate analysis with X, Y, and Z, the relationship between X and Y no longer holds.

    It might also be cool to look at the relationship between number of variables measured, number of individuals measured (sample size), and false-positive rate. If you measure a large number of variables on a small sample of individuals, you are likely to get a false-positive in at least one of those variables being significant (at some designated alpha level).

    Reply

  6. Suggestion 1: restriction of range effects with imperfect filters. So, for example, college admissions cut off the bottom (and for non-CalTech/MIT schools, the top) of the IQ range, but they do so imperfectly. The most naive model is that a typical college cuts off everyone IQ<105, but more realistically, the cutoff goes from 100% excluded at IQ=100 to 0% excluded at IQ=110. How will those sorts of cutoffs affect correlations compared to purely linear restrictions of range?

    Suggestion 0 (because it's logically prior, though I came up with it formulating suggestion 1): How does restriction of range affect the population statistics of the sample in one variable? If I am unable to test people with an IQ under 60 because they're all institutionalized, how does that affect my results? How about IQ<90 because it's an Army sample?

    Reply

  7. Someone really SHOULD tackle that man, because I can’t tell you how he gets communist history wrong, in Russia and the other nations, and I’m not even communist!

    He’s not fond of criticism, and there are some accusations that he took a lot of works from others without representing them correctly. On H-Net he got a lot of criticism, but he brushed them all off.

    I think sometimes, his goals are noble, but on Occidental Observer, he talks about the same things ALL OF THE TIME. It gets boring after a while.

    Reply

  8. rich lawler,

    It sounds like you are talking about a kind of partial correlation scenario where the correlation matrix has controlled for all the other variables. The effect of controlling in this fashion depends not just on the strength of the correlation between two variables, but their actual relationship. So it is not very apt for a visualization.

    As for the second, sample size has no influence on theoretical false positive rates. But sure, one could make this. Let users choose a sample size, number of variables, then it will show a correlation matrix with p values with and without correction for multiple testing. Easy.

    Anthony,

    Imperfect filtering is already planned (basically, I have in mind using a logistic function).

    I will add sample characteristics. I will also add more a more intuitive slider (choose between using Z scores, centiles or raw scores).

    Reply

  9. Hi Chick,

    This comment has nothing to do with the above post, but I thought you’d be interested in the following. At Christie’s in London, in mid-July, a treasure-trove of wonderful old books and manuscripts including Raymond of Penyafort’s (1175-1275) tables of consanguinity and affinity, in Latin is up for auction. It includes a king holding an arrow-like chevron containing a portrait of a man and the blood relatives with whom marriage is prohibited; the table of affinity with a married couple above a panel containing the relatives by marriage who cannot fulfil the role of second spouse upon the death of the first.

    When I got the catalog, I thought of your studies.

    Best to you.

    Reply

  10. @kristen – “At Christie’s in London, in mid-July, a treasure-trove of wonderful old books and manuscripts including Raymond of Penyafort’s (1175-1275) tables of consanguinity and affinity, in Latin is up for auction.”

    oh, how neat! (^_^) thanks so much for sharing!

    i found it on christie’s website: RAYMOND OF PENYAFORT (1175-1275), Quia tractare intendimus, with TABLES OF CONSANGUINITY AND AFFINITY, in Latin, ILLUMINATED MANUSCRIPT ON VELLUM [Paris, mid-13th century].

    Estimate: £30,000 – £50,000 ($47,520 – $79,200).

    *gulp!* wish i had that kind of cash lying around. (~_^)

    Reply

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s