03 April 2011

What's A Blog For, If Not Pet Peeves

Hit and Run at Reason.com points us to a new report on cancer rates in the US:
Trend analysis showed that overall cancer incidence rates for all racial and ethnic groups combined decreased by 0.8% per year during the most recent period, 2003–2007 (Table 1); a statistically significant decrease of 0.6% per year was noted in women, whereas a non-statistically significant decrease of 0.8% per year was noted in men that was influenced by a recent (2005–2007) non-statistically significant increase in prostate cancer incidence. Incidence for prostate and breast cancers, two of the most frequently diagnosed cancers, showed possible changing trends. Cancer of the prostate showed a non-statistically significant annual increase of 3.0% in 2005–2007, after a statistically significant decrease in 2001–2005. The trend analysis of breast cancer in women showed a decrease from 1999 until 2007. However, inspection of the annual breast cancer incidence rates during this period (data not shown) revealed that, after a sharp decrease in rates in 2002–2003, the lower rates subsequently remained stable.
As we've discussed before, the point of statistics is to tell us whether two groups of measurements, which appear to be different, really are different. It looks like more men got prostate cancer last year -- the rate last year is higher than the rate the year before -- but is that a real difference, or is it just chance. We assume that it's chance unless the likelihood of getting these results randomly (given that there really is no difference) is less than 5%.

If the difference is not "statistically significant" then what we're saying is that the difference is, as far as we can tell, effectively zero. There is no difference, except for random chance. So, if the difference is effectively zero, THEN THE FACT THAT THE RATES WERE NOMINALLY DIFFERENT IS COMPLETELY IRRELEVANT. Nothing is more clearly indicative of someone trying to use statistics to lie to you then telling you the direction of a non-significant change.

5 comments:

erp said...

The purpose of reports is to validate whatever the government propaganda mill wants us to believe.

They're about to start rationing treatment, so they're softening us up by spreading the good news that less treatment will be necessary in the future.

Statistics are just an easy vehicle to get the message across because so few people understand them.

Bret said...

I disagree.

The 95% confidence number is arbitrary. The trends are certainly "significant" using some "confidence" level.

Harry Eagar said...

Confusion abounds between statistical significance and meaningful difference.

That's a pretty subtle softening up maneuver, erp.

Of course, under the Republicans, everybody got all the health care necessary. There were never any complaints.

erp said...

Harry, it's about as subtle as sledgehammer and Republicans aren't about control, the Dems are and they sure proved since Obama got elected.

Pet Lover said...

These numbers and stats are just an eyewash to the public.