0
$\begingroup$

I want to ask if I can use the NormalPValue[val] the same way for lower-tail and upper-tail hypothesis tests or if there is something I need to watch out for? Personally, my gut feeling is that I need to substract the P-Value from 1 for the other kind of test.

Here's an example to show what I'm talking about:

A software reduced the amount of spam of 100 messages to 11%. This is better than the goal of 20%. Based on a significance level of 0.05, is this significant or not? I.e. buy the software or not?

  1. We state the following hypotheses:

H0: p >= p0 = 0.20

H1: p < p0 = 0.20

Further use [Alpha] = 0.05.

  1. Since we test a proportion we have to use the z-statistic:

    phead = 0.11;

    p0 = 0.20;

    n = 100;

    zcalc = (phead - p0) / (Sqrt[p0 * (1 - p0) / n])

    -2.25

  2. Calculate the P-Value Needs["HypothesisTesting`"];

    NormalPValue[zcalc]

    OneSidedPValue -> 0.0122245

  3. Compare P-Value to [Alpha]

    Since The P-Value of 0.0122245 is smaller than our [Alpha] = 0.05, we reject H0 and accept the alternative hypothesis H1. The manager should buy the software.

My question now is, if we were to use the following H0/H1 instead (for a different kind of task, of course):

H0: p <= p0 = 0.20

H1: p > p0 = 0.20

Am I correct that in this case, I would receive the P-Value as follows:

1 - 0.0122245 = 0.9877755

And would therefore not be able to reject H0?

|improve this question
$\endgroup$
  • $\begingroup$ This site is about the programming environment "Mathematica" not about mathematics and statistics in general. You might want to take a gander at crossvalidated.SE for questions about statistics. $\endgroup$ – Sascha Jan 18 '17 at 12:54
  • $\begingroup$ @Sascha My question stems from Mathematica, since unlike when using TTest (reference.wolfram.com/language/ref/TTest.html) there is no option to define whether it's a lower-tail or upper-tail test in the OneSidedPValue[val] function. I.e. I don't understand what OneSidedPValue returns and how to properly use it. $\endgroup$ – Jonas Kaufmann Jan 18 '17 at 12:58
  • $\begingroup$ You might want to edit your question to reflect what you stated in your comment. $\endgroup$ – Sascha Jan 18 '17 at 13:00
  • $\begingroup$ There are some issues with the example you've used. The distribution that is assumed is a binomial distribution and the "exact" answer for the first set of hypotheses is CDF[BinomialDistribution[100, 0.2], 11]. For the second set of hypotheses the answer is 1 - CDF[BinomialDistribution[100, 0.1], 10] which is not "1 minus the first probability". A "P-value" is the probability of obtaining at least as extreme a value as observed. (I've ignored the complication that you have a compound hypothesis for the null hypothesis.) $\endgroup$ – JimB Mar 20 '17 at 0:07
  • $\begingroup$ If your underlying distribution was continuous, then, yes, the P-values would be 1 minus the other. Also, in that case you could use LocationTest which does have an option for a one-sided alternative hypothesis. $\endgroup$ – JimB Mar 20 '17 at 0:09
1
$\begingroup$

To answer my own question (thanks to some help from a colleague in real life): Yes, it is correct, since we need to get the opposite area of the normal distribution area.

|improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.