I am running a test where a group of individuals will sample 2 products and
provide a preference for one or the other, as well as an overall rating for
each product.
Okay - what test will tell me a) that the preference rating indicates a
"true win" for one product or the other?; and, b) that the average overall
ratings for the 2 products are or are not significantly different?
I have some ideas, but would appreciate any help that's out there!!! THANKS!

Just do a count of preference results
do you prefer product A or B
count the A's and B's and that tells you which product "wins"
then do a scale such as 1-10 to rate each, and take an average for each
product.

--
--
-John
Please rate when your question is answered to help us and others know what
is helpful.

"Pizza" wrote in message
...
I am running a test where a group of individuals will sample 2 products and
provide a preference for one or the other, as well as an overall rating
for
each product.
Okay - what test will tell me a) that the preference rating indicates a
"true win" for one product or the other?; and, b) that the average overall
ratings for the 2 products are or are not significantly different?
I have some ideas, but would appreciate any help that's out there!!!
THANKS!

Generally speaking, you need to:
Decide what your testers are going to test for, and how they will score it. For example, if you
were testing Pizza, you might have the panelists test the crust for lightness, crispness,
doughiness, etc; the sauce for saltiness, tomato flavor, spices; the topping for greasiness, taste,
stringiness....

You may want to train the panelists on how to differentiate between the qualities that you are
looking for.

1) Keep your testers in the dark about how many different products they are testing
2) Randomly present the products to the individials - sometimes product 1 first, other times product
2 first, and strive to keep the products consistent in their presentation
3) Repeat your tests, so that each individual "tests" each product multiple times
4) Hire a PhD Statistician......err,
4a) Look at the mean, variance, spread, standard deviation of your data, based on overall, each
judge, order presented, and over time. Look for trends, and then you need to decide what is
meaningful. For example, you could decide that the mean values need to differ by at least the
largest of the standard deviations, the mean values need to differ by the sum of the two different
standard deviations, etc.

Remember, there are liars, da*ned liars, and statisticians.

HTH,
Bernie
MS Excel MVP


"Pizza" wrote in message
...
I am running a test where a group of individuals will sample 2 products and
provide a preference for one or the other, as well as an overall rating for
each product.
Okay - what test will tell me a) that the preference rating indicates a
"true win" for one product or the other?; and, b) that the average overall
ratings for the 2 products are or are not significantly different?
I have some ideas, but would appreciate any help that's out there!!! THANKS!

The test plan is clear - but the question is...when I get back all the
responses to "Which Pizza crust did you prefer - A or B - and I find that 78
of 100 people say A, is there a statistical test that tells me how safely I
can state that 78% of all pizza eaters will prefer A? Confidence interval?
For comparing the ratings - do I use a t-test (independent or paired?). And
also hire the PhD Statistician! Have you read "How To Lie With Statistics", a
cute little book written back in the 50's I think. Oh yeah, plenty o' dem
liars! And plenty o' ways to lie!
Thanks for your interest and feedback.


"Bernie Deitrick" wrote:

Generally speaking, you need to:
Decide what your testers are going to test for, and how they will score it. For example, if you
were testing Pizza, you might have the panelists test the crust for lightness, crispness,
doughiness, etc; the sauce for saltiness, tomato flavor, spices; the topping for greasiness, taste,
stringiness....

You may want to train the panelists on how to differentiate between the qualities that you are
looking for.

1) Keep your testers in the dark about how many different products they are testing
2) Randomly present the products to the individials - sometimes product 1 first, other times product
2 first, and strive to keep the products consistent in their presentation
3) Repeat your tests, so that each individual "tests" each product multiple times
4) Hire a PhD Statistician......err,
4a) Look at the mean, variance, spread, standard deviation of your data, based on overall, each
judge, order presented, and over time. Look for trends, and then you need to decide what is
meaningful. For example, you could decide that the mean values need to differ by at least the
largest of the standard deviations, the mean values need to differ by the sum of the two different
standard deviations, etc.

Remember, there are liars, da*ned liars, and statisticians.

HTH,
Bernie
MS Excel MVP


"Pizza" wrote in message
...
I am running a test where a group of individuals will sample 2 products and
provide a preference for one or the other, as well as an overall rating for
each product.
Okay - what test will tell me a) that the preference rating indicates a
"true win" for one product or the other?; and, b) that the average overall
ratings for the 2 products are or are not significantly different?
I have some ideas, but would appreciate any help that's out there!!! THANKS!

Statistics are bad at best when using a small sample, go to Alabama and
taste test some chitlins' and you might get 15% or so that like them, do
that in New York and get beaten. There are lots of formulas, but to get any
kind of real valid data you need a LARGE sample. To help the lying have them
test each twice, and determine if their answers were similar. Wonder what
the statistics are of people who are willing to participate versus those who
are not, you can never know, but you can state what your sample background
is, 35% women 65% Men 25% blondes, whatever is relavent, then you can
reasonably say that 78% of people with a similar background as the sample
would buy the product.

--
--
-John
Please rate when your question is answered to help us and others know what
is helpful.

"Pizza" wrote in message
...
The test plan is clear - but the question is...when I get back all the
responses to "Which Pizza crust did you prefer - A or B - and I find that
78
of 100 people say A, is there a statistical test that tells me how safely
I
can state that 78% of all pizza eaters will prefer A? Confidence
interval?
For comparing the ratings - do I use a t-test (independent or paired?).
And
also hire the PhD Statistician! Have you read "How To Lie With
Statistics", a
cute little book written back in the 50's I think. Oh yeah, plenty o' dem
liars! And plenty o' ways to lie!
Thanks for your interest and feedback.


"Bernie Deitrick" wrote:

Generally speaking, you need to:
Decide what your testers are going to test for, and how they will score
it. For example, if you
were testing Pizza, you might have the panelists test the crust for
lightness, crispness,
doughiness, etc; the sauce for saltiness, tomato flavor, spices; the
topping for greasiness, taste,
stringiness....

You may want to train the panelists on how to differentiate between the
qualities that you are
looking for.

1) Keep your testers in the dark about how many different products they
are testing
2) Randomly present the products to the individials - sometimes product 1
first, other times product
2 first, and strive to keep the products consistent in their presentation
3) Repeat your tests, so that each individual "tests" each product
multiple times
4) Hire a PhD Statistician......err,
4a) Look at the mean, variance, spread, standard deviation of your data,
based on overall, each
judge, order presented, and over time. Look for trends, and then you
need to decide what is
meaningful. For example, you could decide that the mean values need to
differ by at least the
largest of the standard deviations, the mean values need to differ by
the sum of the two different
standard deviations, etc.

Remember, there are liars, da*ned liars, and statisticians.

HTH,
Bernie
MS Excel MVP


"Pizza" wrote in message
...
I am running a test where a group of individuals will sample 2 products
and
provide a preference for one or the other, as well as an overall rating
for
each product.
Okay - what test will tell me a) that the preference rating indicates a
"true win" for one product or the other?; and, b) that the average
overall
ratings for the 2 products are or are not significantly different?
I have some ideas, but would appreciate any help that's out there!!!
THANKS!

Pizza -

If your sample is randomly selected, you can construct a confidence
interval. Browse to google.com and search for "confidence interval
proportion' (without the quotes).

If your sample is not a probability (usually simple random) sample, then you
cannot construct a confidence interval.

Your samples seem to be independent (unless you have specified some pairing
method), and for relatively large sample sizes (once again necessarily
randomly selected) you could use the normal approximations for a "hypothesis
test of the difference between two population proportions." Consult a basic
statistical inference textbook or google.

- Mike
http://www.mikemiddleton.com

"Pizza" wrote in message
...
The test plan is clear - but the question is...when I get back all the
responses to "Which Pizza crust did you prefer - A or B - and I find that
78
of 100 people say A, is there a statistical test that tells me how safely
I
can state that 78% of all pizza eaters will prefer A? Confidence
interval?
For comparing the ratings - do I use a t-test (independent or paired?).
And
also hire the PhD Statistician! Have you read "How To Lie With
Statistics", a
cute little book written back in the 50's I think. Oh yeah, plenty o' dem
liars! And plenty o' ways to lie!
Thanks for your interest and feedback.


"Bernie Deitrick" wrote:

Generally speaking, you need to:
Decide what your testers are going to test for, and how they will score
it. For example, if you
were testing Pizza, you might have the panelists test the crust for
lightness, crispness,
doughiness, etc; the sauce for saltiness, tomato flavor, spices; the
topping for greasiness, taste,
stringiness....

You may want to train the panelists on how to differentiate between the
qualities that you are
looking for.

1) Keep your testers in the dark about how many different products they
are testing
2) Randomly present the products to the individials - sometimes product 1
first, other times product
2 first, and strive to keep the products consistent in their presentation
3) Repeat your tests, so that each individual "tests" each product
multiple times
4) Hire a PhD Statistician......err,
4a) Look at the mean, variance, spread, standard deviation of your data,
based on overall, each
judge, order presented, and over time. Look for trends, and then you
need to decide what is
meaningful. For example, you could decide that the mean values need to
differ by at least the
largest of the standard deviations, the mean values need to differ by
the sum of the two different
standard deviations, etc.

Remember, there are liars, da*ned liars, and statisticians.

HTH,
Bernie
MS Excel MVP


"Pizza" wrote in message
...
I am running a test where a group of individuals will sample 2 products
and
provide a preference for one or the other, as well as an overall rating
for
each product.
Okay - what test will tell me a) that the preference rating indicates a
"true win" for one product or the other?; and, b) that the average
overall
ratings for the 2 products are or are not significantly different?
I have some ideas, but would appreciate any help that's out there!!!
THANKS!