What statistical calculations could i use? ?
im doing something on statistics- comparing males and females at something. im trying to find whose estimated were more accurate- male or female. ive used standard deviation, mode, median, mean...none of that is actually answering my question. i cant use correlation coeficient (or can i?) because im comparing a set of numbers to a category (male/female) what else can i do to answer my question? help!! please im desparate? could i use chi2?
and whats the confidence interval? i dont really care if i get reliable answers, i have 15 males/females. its a school project, i cant have 10 000 haha.
- ??????Lv 71 decade agoFavorite Answer
What you really need is a sample size given a
certain confidence interval.
You need to know if there is a significant difference
between the estimates of males and females.
Take for instance 95 % confidence and 1 %
deviation, then you need to know how many
samples you need to take to have a significant
different number. I can tell you that sample sizes
can be tremendously big ! You typically have to
take 10,000 samples with 95% confidence and
1 % tolerated deviation.
The sample size can be calculated exactly with
a C program or roughly with Z - value.
The point is if you only have a few hundreds of
sample data, you cannot be certain about the
difference between males and females !!!!!!
Take for instance a coin toss. You would want to find
out if a coin is fair by throwing only 15 times. This is
not possible ! Confidence is that you are e.g. 95 %
certain that the real P[head] is between
[outcome - tolerated deviation, outcome + tolerated diff.]
But for 95 % (usual taken value) you need to throw
thousands of times with 1% tolerated deviation !
- Anonymous1 decade ago
Honestly, the easiest thing you can do is take a confidence interval. It depends on exactly what you are measuring (different things can be expected to have different distributions), but often a simple confidence interval using the t-distribution, or even the normal distribution, will suffice. The category that has the confidence interval with the smallest width is that for which the estimate is most accurate.
There are a number of other methods you can use - you can for example use an F-test to determine whether the standard deviations of the two samples are statistically different from each other. I think that the confidence interval approach is however sufficient for this.