d******s
发帖数: 231 | 1 Hope statistians could help me... apologies if i am not speaking the right
statistics language.
For example UPS ground takes an average of Q days for delivery. Priority
mail takes an average of Q+1 days for delivery. I want to know with N
measurements, in order to have 95% confidence to say Priority mail is 1 day
slower than UPS, how much sigma (std dev) should I have in the measurements,
assuming variances are the same?
Thanks! | x***x
发帖数: 3401 | 2 you can do a paired t-test and say the 95% confidence interval of the difference of
the avg delievery time between UPS and USPS is (x, y). If your confidence
interval is indeed above 1 (i.e, x>1), you can conclude usps is significantly at least 1
day slower than ups.
the minimum std dev is dependent on the avg difference in your N measurement
. | l*********s
发帖数: 5409 | 3 normality assumption is of questions.
difference of
significantly at least 1
measurement
【在 x***x 的大作中提到】
: you can do a paired t-test and say the 95% confidence interval of the difference of
: the avg delievery time between UPS and USPS is (x, y). If your confidence
: interval is indeed above 1 (i.e, x>1), you can conclude usps is significantly at least 1
: day slower than ups.
: the minimum std dev is dependent on the avg difference in your N measurement
: .
| x***x
发帖数: 3401 | 4 valid under large N (e.g. N>30)
【在 l*********s 的大作中提到】
: normality assumption is of questions.
:
: difference of
: significantly at least 1
: measurement
| s*r
发帖数: 2757 | 5 why paired t-test
why "confidence interval is indeed above 1 is equivalent" to "x>1"
difference of
significantly at least 1
measurement
【在 x***x 的大作中提到】
: you can do a paired t-test and say the 95% confidence interval of the difference of
: the avg delievery time between UPS and USPS is (x, y). If your confidence
: interval is indeed above 1 (i.e, x>1), you can conclude usps is significantly at least 1
: day slower than ups.
: the minimum std dev is dependent on the avg difference in your N measurement
: .
| g**r
发帖数: 425 | 6 paired t就可以 control其他的 noise了吧，比如周一发的比周日发的可能快。
如果楼主不管这些细节，普通t就行了。
(x,y)是CI，x>1的那个人CI>1。
楼主对CI的表述不太准确,说with 95% confidence...， 如果细究起来，CI不能这样理
解啊。
【在 s*r 的大作中提到】
: why paired t-test
: why "confidence interval is indeed above 1 is equivalent" to "x>1"
:
: difference of
: significantly at least 1
: measurement
| x***x
发帖数: 3401 | 7 因为delivery time is highly correlated with delivery distance. your test
should control the delivery distance factor.
我想的是至少sample两两的original zip code和destination zipcode得一样
比如一个从NYC到Chicago, 一个USPS的sample 一个UPS的sample
从Atlanta到LA, 一个USPS的sample, 一个UPS的sample
如果USPS和UPS的delivery distance都对不上的话, avg delivery time就很容易biased 结论也
就没有意义了 当然就像LS同学说的 不要这么严谨的话 simple t-test就可以了
x是CI的lower bound, 如果x>1, 那自然整个CI都在1的右边啊.
【在 s*r 的大作中提到】
: why paired t-test
: why "confidence interval is indeed above 1 is equivalent" to "x>1"
:
: difference of
: significantly at least 1
: measurement
| s*r
发帖数: 2757 | 8 I did not realize the (x,y) is a CI
ups divides US into several zones, and gives similar delivery time line
within zones. it seems not that sensitive to distance,but still good
argument for a paired test
biased 结论也
【在 x***x 的大作中提到】
: 因为delivery time is highly correlated with delivery distance. your test
: should control the delivery distance factor.
: 我想的是至少sample两两的original zip code和destination zipcode得一样
: 比如一个从NYC到Chicago, 一个USPS的sample 一个UPS的sample
: 从Atlanta到LA, 一个USPS的sample, 一个UPS的sample
: 如果USPS和UPS的delivery distance都对不上的话, avg delivery time就很容易biased 结论也
: 就没有意义了 当然就像LS同学说的 不要这么严谨的话 simple t-test就可以了
: x是CI的lower bound, 如果x>1, 那自然整个CI都在1的右边啊.
|
|
