I need help with interpretation of P-value please

#1
Hello

I need anyone who understands these things to help with interpretation of an output from minitab please.

I am looking at whether results for a test (INR) from a handheld device are statistically different from a result from the same patient processed at a nationally accredited lab.

Two-Sample T-Test and CI: Lab INR, POCT INR

Two-sample T for Lab INR vs POCT INR

_ _ _ _ __ N __ Mean _ StDev _ SE Mean
Lab INR __ 58 _ 2.576 _ 0.785 __ 0.10
POCT INR_ 58 _ 2.429 _ 0.831 __ 0.11
(Underscores added to format the HTML better)

Difference = mu (Lab INR) - mu (POCT INR)
Estimate for difference: 0.146552
95% CI for difference: (-0.150828, 0.443931)
T-Test of difference = 0 (vs not =): T-Value = 0.98 P-Value = 0.331 DF = 113

I understand it satisfactorily up to the interpretation of the P-value.
Do I say that my null hypothesis (handheld gives same result as lab) is not rejected because the P-Value is within the range of the 95% Confidence Intervals?
Or do I reject the null hypothesis because the P-value is greater than my alpha of 0.05?

Thanks for any help

Ben Quinn.
 

Lark

New Member
#2
Ben,

You accept the null hypothesis because the p value is not that unusual (ie it is not below 0.05).

You can say: The difference in the means is only significant at the .33 level. (this may not be the best wording, I don't have time to look this up right now, maybe someone else can help here) (It is less precise to say that the "difference in the means is not significant at the 0.05 level"; but of course this is true).

The 95% confidence interval for the difference in the means is (-.15, .44). Ie, you can't say there is a difference in the means with a 95% confidence because the difference could be 0. 0 is in the interval.

P values are often easy to misinterprete from software; one way to figure it out is to put some large differences in means and see how this affects the p value.

Lark
 
#3
P-Values

Lark

Thanks for that. Would this be right then - a p-value of 0.331 would be significant with an alpha of >0.331?

Also, this is a two-tailed test - the mean of the handheld tests could have fallen either side of the mean of the lab tests - if we are using an alpha of 0.05, do we halve this for each tail and say p > 0.025 & do not reject null hypothesis (for 95% confidence limits)?

Ben.
 

Lark

New Member
#4
Ben,

I looked in some of my stat books and I am not comfortable with your wording. I would stick with my wording or say " that there is no statistical difference in the means, this is at a significance level of p= 0.33). I think you could say this is at an alpha of 0.33. By the way: how do you make an alpha on the keyboard?

I calculated the p value myself (see the formuala for t test in any stat book) and the value of .33 is from a 2 sided test. If it was from a 1 sided test then you would have to double the p value if you were to report it as a 2 sided test.


Lark
 
#5
Thank you.

Lark

Sorry for the delay replying - been in Ireland and statistics were the furthest thing from my mind.
Thanks for your advice.

I can make an "alpha" in Word using Insert|Symbol from the toolbar. Select the basic greek character set and browse to the alpha symbol. You can use the same dialogue box to assign a shortkey if you want to use it frequently.

Thanks for your help.

Ben.