Terry

Marc wrote:
> gRmEcMgOrVeEw@mindspring.com (Gordon McGrew) wrote:
>
>>On Tue, 25 Mar 2003 19:45:25 -0900, Marc <whineryy@yifan.net> wrote:
>>
>>>lparker@emory.edu (Lloyd Parker) wrote:
>>>
>>>>In article <v7vnt85u26osec@corp.supernews.com>,
>>>> "Floyd Rogers" <floydr@accessone.com> wrote:
>>>>
>>>>>"Gordon McGrew" <gRmEcMgOrVeEw@mindspring.com> wrote
>>>>>
>>>>>>>>Look up "anecdotal" and get back to us.
>>>>>>>
>>>>>>>Anecdoatal (adj.): See "Consumer Reports."
>>>>>>
>>>>>>In what way is the CR reliability survey anecdotal?
>>>>>
>>>>>In several ways, but the most important are:
>>>>>1) CR subscribers are not a random sample
>>>>
>>>>They are a representative sample though. A random sample is good for
>>>>eliminating biases, not not an absolute requirement.
>>>
>>>What biases? A random sample is unnecessary if you can determine how your
>>>sample differs from the norm. How does CR account for this? They presume
>>>no biases in their readership or those that return the surveys.
>>
>>I think that is a reasonable presumption. The responses are based on
>>actual experience, not opinion. If your transmission failed, you
>>check the box.
>
>
> Is the **** on the shifter falling off a transmission failure? How about
> if the person caused the failure by improperly shifting, but blames it on
> the transmission when asked in the survey?

Marc... I'm start'n to see yer point. A much higher percentage of the
people who buy domestic vehicles are likely to be dumb enough to damage
their transmission by shifting improperly than would the average buyer
of Japanese vehicles. Therefore, no meaningful surveys can be made
unless there is first a survey taken to ascertain the level of stupidity
in the respective groups. Brilliant!! I'm on yer side, bud.

>
>
>>Another way of looking at it is that, since they surveyed CR readers,
>>then the results are valid for anyone who reads them.
>
>
> Then every one here would appreciate you not telling us what is reliable
> based off CR, as the results may be valid within the self selected group
> (you) and not outside (the rest of the world).
>
>
>>>CR apparently doesn't correct for the number sold. If there are 1,000,000
>>>Ford Excretions sold, and they have data on 1000 of them, they consider
>>>that as valid and representative as 1000 responses from the SUX 6000, of
>>>which only 1000 were made.
>>
>>CR assumes that the actual percentage of vehicles surveyed is
>>insignificant. 500 or 5000 is sufficient to report results regardless
>>of how large the population is. OTOH, if they got 50 reports on a
>>vehicle that sold only 100, they would presumably throw it out even
>>though it would be an adequate sample. (This is hypothetical since
>>the chance of such a thing happening is negligible.)
>
>
> Then you see my point. They may have a statistically valid sample and
> throw it out because it doesn't meet their proprietary (non-scientific)
> criterion. Yet, they would keep an invalid sample of a large production
> run because it meets the proprietary (non-scientific) criterion.
>
>
>>>The same goes for the number for an
>>>"insufficient data" response. The number of vehicles made should impact
>>>the validity of the data, but CR doesn't account.
>>
>>As explained in another post, the n affects the confidence interval
>>but it doesn't affect validity. Obviously CR could provide a huge
>>data dump from this survey but the point is to present the results in
>>a concise, easy-to-understand form.
>
>
> They have simplified it in such a way that if the rest of the process was
> properly scientific (and it is not), that the end results do not meet the
> requirements. They need to present the confidence. Without reporting the
> number of samples and the population size, they are not meeting this
> requirement.
>
> It is scientifically invalid. It fails on *many* levels. It fails in
> gathering data. It fails in processing the data. It fails in the
> presentation of the data. There isn't a single point at which it is
> handled in a manner designed to be a scientific survey.
>
> It may be useful. It is not scientifically valid.
>
> Marc
> For email, remove the first "y" of "whineryy"