Lies, damn lies, and statistics

[Brian Howell]

Oops! It seems that a lone 19-year-old male African-American Illinoian has been messing up the L.A. Times' polling.

It seems they were using weighted sampling, which is a common statistical technique, to achieve demographic (or other) proportionality within an otherwise biased sample. However, in this case, the demographic significance (rarity) of the individual in question resulted in him being weighted from thirty to three hundred times other respondents.

The foregoing is a roundabout way of getting to my real question: it is increasingly difficult for political pollsters to obtain a sample, leastwise a representative cross-sectional one. So, are polls still meaningful? Do they still have real value?

http://www.slate.com/blogs/the_slatest/2016/10/12/one_of_trump_s_african_americans_in_illinois_is_messing_up_the_l_a_times.html

[jack saunders]

On Oct 12, 2016, at 2:46 PM, Brian Howell 

Polls have declined in predictive value because anyone who picks up a landline when it rings is, ipso facto, six standard deviations from any norm.

The demographic fudge factors you mention are the inexpensive work-around.  They have yielded an unusual number of whopping errors over the last decade...oddly in line with monster hurricanes.  So the standard caution about causation is well taken.  But mass media commercial polls have no predictive value.  If very wealthy interests deeply care, they commission hand-made polls.....which is exactly what it seems: button-hole a truly represntative sample and ask them questions.  Only giant media and the Koch bros (and probably Wells Fargo, about now) can afford this.

--
You received this message because you are subscribed to the Google Groups "Ipse Dixit" group.
To unsubscribe from this group and stop receiving emails from it, send an email to Ipse-dixit+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/Ipse-dixit/da2ea2de-8da9-488c-bc5c-a5c9e685acda%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.