TTest in Excel?

Question

Okay im trying to do a TTest using excel and i am lost.

I have Dog Whelk specimens from 2 different communities that I am comparing. They were collected from two different places. (One where there where heavy waves and the other from a more sheltered place.)

I took measurements (length,width, and aperture) of 30 members of each community.

I need to see if there is a signifigancet difference between these two commuinites.

I put the data in and click on the ttest choose which things i wanted to compare and this is where i am confused.

I dont know what tails and type to use.

I am assuming the "paired(1)" for type but im not sure if that is right either.

But now im lost. I dont know what to put for tails.

I need to see if the P-value i get is grater or less than .05.

Answer 1

You probably don't want to use "paired" samples.  This is when you actually have pairs of samples that each have some similar thing acting on them.  Like for example, does one cricket chirp faster than another.  If you have two crickets and you record them on different nights, and if their chirping is temperature dependent (it is for some crickets), then you would pair up the crickets for each night--basically you are just looking at the difference between their times rather than the total averages of their times together.  like 10&11, 50&51, 5&6, 100&101.  Each has a difference of 1, which is significant when you pair them.  But if you look at two groups (10, 100, 50, 5) and (101, 6, 51, 11), the variance is so great and the diference in means is so small that you will not have  a significan difference.  Only put in pairs if you really have a good reason to, because you cut your effective sample number in half!  From your description it doesn't sound paired.

Your test depends on your hypothesis.  If your hypothesis is that "group A is bigger than group B", then you use a one sided test.  It gives you a better chance at significance, but it doesn't tell you anything if A is less than B.  If your hypothesis is just that "group A doesn't equal group B" then you use two sided tails, which allows you to test either greater than or less than--giving you more flexibility in your final answer, but it also gives you a larger chance of not rejecting the null hypothesis--i.e. better chance that p value is greater than .05.  

you are supposed to make the decision of whether to do a one-tailed test or two-tailed test before you do the test.  Otherwise, you are tempted to look at your result and decide on the the test afterwards, which biases your result.  Basically, if your hypothesis is waves make smaller whelks, then you do a one sided test, and stick with your null hypothesis even if the opposite of what you predicted happens.  If your hypothesis is just that waves affect the growth of whelks, then do a two sided test and in the end state whether it increased or decreased their growth or if there was no statistical difference.  

Hope that helps and good luck!

Answer 2

'Paired' is wrong. I do not know where you get the 1 from.

Do three t -tests.

e.g

1) compare lengths_a with lengths_b to test a hypothetised difference between the two population means. Enter a 0 for the difference tested for. The safest t-test to use in this context assumes as little as possible i.e. t-test two sample, assuming unequal variance. The hypothetised mean difference could be zero.

Look at the two-tail results from Excel. In my own pretend scenario I got t (calculated) = - 0.249 and a critical t value of +2.179. That result would suggest no significant difference between the average population lengths of the two populations. I used alpha = 0.05 as you have to do. Excel uses alpha = 0.05 by default. In my case P(T<=t) two-tail

= 0.808 when it would need to be < 0.05 for a statistically significant difference to exist.

Here is my raw data (invented)

1,1.2

2,1.5

3,3.9

4,3.9

5,4.8

6,6.7

7,8.2

for which I could come to the conclusion I gave you.

t-Test:

Here is the Excel output:

Two-Sample Assuming Unequal Variances



Variable 1 Variable 2

Mean 4 4.314285714

Variance 4.666666667 6.498095238

Observations 7 7

Hypothesized Mean Difference 0

df 12

t Stat -0.248856465

P(T<=t) one-tail 0.403839991

t Critical one-tail 1.782286745

P(T<=t) two-tail 0.807679981

t Critical two-tail 2.178812792