click on LOGO to return to HOME page

"Our job is to give people not what they want, but what we decide they ought to have."---Richard Salant (former president of CBS news)

Criminal offense?
Imagine a corporation using the following methods. Creative math sends corporate executives to jail.
Here it is used to justify extreme laws and loss of freedom affecting all of us based on "simulated", "missing" & "unknown" numbers.

A Little Known Fact: NHTSA assigns Blood Alcohol Content (BAC) values to 60% of the drivers who the police felt no need to test for alcohol.
Doesn't it seem strange that in a fatal accident a driver would not be tested if the police had any suspicion that he may be drunk?

If 60% of the so-called drunk drivers were never even tested, how accurate can their statistics be? Garbage In, Garbage Out.

The Government 'Fesses Up:

Here is the government's exact wording (isn't bureaucratic language fun?):

of alcohol involvement in fatal crashes for the U.S. are based on data from NHTSA's Fatality Analysis Reporting System (FARS).   Known BAC [Blood Alcohol Content] test results are not available for all drivers and non-occupants involved in fatal crashes for a number of reasons, most frequent of which is that persons are not always tested for alcoholTo address missing data, NHTSA has developed and employs a statistical model to estimate the likelihood that a fatal crash involved driver or non-occupant was sober (zero BAC), had some alcohol (BAC of 0.01-0.09) or was intoxicated (BAC of 0.10) at the time of the crash.  The statistical model was developed using all available known data in the aggregate (that is, at the national level) and applied to each individual driver and non-occupant with an unknown BAC test result. The estimates include a mix of both known and estimated BACs."---(DOT HS 809 334). [Emphasis added].

Translation: 60% of the drivers were not tested. Why? Maybe there was no reason to test them. So they take a guess. They assume, and we all know what assuming is: making an ASS out of U and ME.
This disclaimer is hidden deep in the bowels of the NHTSA web site:

"It is necessary to emphasize that none of the tabulations presented can be interpreted as implying a direct causal relationship between alcohol use and any other attribute of fatal crashes. Inferences concerning causality can only be made on the basis of additional data that is independent of the FARS data
." [FARS = Fatal Accident Analysis Reporting System--U.S. government data arm of the NHTSA]

Translation: The presence of alcohol may or may not have caused the crash. They don't know and apparently don't intend to try and find out.

More quotes from the NHTSA:
"The new method estimates BAC levels over the entire range of plausible values from 0.00 to 0.94 g/dl."
Since you are probably DEAD at .45 g/dl (BAC level), doesn't the inclusion of .46-.94 "plausible values" skew the statistical results? Duh!
More Confessions on how they make up the numbers:

Verbatim text found in an obscure file at:  National Center for Statistics and Analysis, 400 Seventh St., S.W., Washington, DC 20590--This is a branch of the NHTSA. Some answers have been deleted for brevity. The questions were left to assure you that no salient information has been omitted. "Multiple Imputation" is the government's name for their method of filling in missing data.
[Emphasis in bold print, nothing else was added or changed.]

Appendix A: FAQ on the Multiply-Imputed Datasets of Missing BAC in FARS
[Fatality Analysis Reporting System
, another branch of NHTSA]

1. What is imputation?

A. Imputation is the practice of ‘filling in’ missing data with plausible values. It solves

the missing-data problem at the beginning of the analysis.

2. Why impute Missing BAC in FARS?

A. On an average, approximately 60 percent of the BAC values are missing/unknown in

FARS each year. Invalid inferences can be drawn on the level of alcohol

involvement for cases where the BAC is missing as the characteristics of the persons

with unknown BACs can be significantly different from those with known BACs. In

order to perform complete-data analysis of FARS data with respect to alcohol

involvement, the missing BACs need to be simulated (imputation!)

3. What is Multiple Imputation (MI)?

A. MI is a technique in which each missing value is replaced by m>1 simulated versions

and these simulated complete datasets are analyzed by standard methods. These

simulated values are actual values of BAC in the plausible range (.00<=BAC<=.94).

4. Why Multiple Imputation of BAC in FARS?

A. Multiple Imputation is the state-of-the-art technique to impute missing values. Each

missing BAC value is replaced by ten simulated values of BAC using rigorous

statistical techniques that consider the interaction of all the characteristics of the case.

MI allows for the computation of Standard Errors and Confidence Intervals.

5. Can MI estimates be used in analysis (regression etc.)?

6. How do I combine the results across the multiply imputed datasets?

7. Why not just impute once?

8. Will the alcohol involvement estimates change from those of the previous method?

A. Yes, there will be minor differences between the estimates of alcohol involvement

between the earlier method (Discriminant Analysis) and Multiple Imputation. The

MI estimates are overall between 0 to 2 percent higher than the estimates from the old


9. Why are there differences between the results from the two methods?

10. Are there sample programs that analyze the multiply imputed datasets?

A study by the NHTSA, titled, Alcohol Involvement in Fatal Crashes--Comparisons among Countries, concluded (in their own words):

"The results of the inquiry indicate that the definitions used in the United States to track alcohol involvement in fatal crashes are not shared by other developed countries."

"I could prove God statistically."---George Gallup



Hit Counter