The RapidMiner community is on read-only mode until further notice. Technical support via cases will continue to work as is. For any urgent licensing related requests from Students/Faculty members, please use the Altair academic forum here.

# T-Test

Member Posts: 39 Maven
edited November 2018 in Help
Hi,

what type of significance test is done by the T-Test operator (left-sided, right-sided or both-sided)? I couldn't find any info in the documentation.

Thanks

• Member Posts: 347 Maven
Hello,

the description says:
 Determines if the null hypothesis (all actual mean values are the same) holds for the input performance vectors.
so I say it is a both-sided test with H0: mu1=mu2, H1: mu1!=mu2

greetings

Steffen
• Member Posts: 39 Maven
steffen wrote:

Hello,

the description says:
so I say it is a both-sided test with H0: mu1=mu2, H1: mu1!=mu2

greetings

Steffen
I looked at the source code and it does look like a left sided test, doesn't it?
• Member Posts: 347 Maven
Nope

As far as I see the following method TTestSignificanceTestOperator#getProbability(PerformanceCriterion pc1, PerformanceCriterion pc2) calculates the p-value of the test. The test itself is ''performed'' here:
TTestSignificanceTestOperator#TTestSignificanceTestResult#toString() , i.e. here:
`result.append("Values smaller than alpha=" + Tools.formatNumber(alpha) + " indicate a probably significant difference between the mean values!" + Tools.getLineSeparator());`
the comparison of pvalue < alpha is no clear indication for a left-sided test...

this is the used formula:
http://en.wikipedia.org/wiki/Student';s_t-test#Unequal_sample_sizes.2C_equal_variance => two-sided-test

BUT looking at this formula raises another questions:
First: I have read somewhere that the assumption of equal variances is not a problem if the sizes of the test samples are equal. On the other hand, if this is not valid, no one can guarantee anything for the true alpha error. What do you think about it?

Second:
I thought in case of  a two-sided test the alpha parameter must be divided by 2. Or is this already implied by the test statistics ?

greetings

Steffen

PS:

I looked at the source code and it does look like a left sided test, doesn't it?
I prefer such argumentations with class names and line numbers

• Member Posts: 39 Maven
steffen wrote:

Second:
I thought in case of  a two-sided test the alpha parameter must be divided by 2. Or is this already implied by the test statistics ?
That's why I guessd it's a one sided test.
• Member Posts: 347 Maven
Hello again
steffen wrote:

Second:
I thought in case of  a two-sided test the alpha parameter must be divided by 2. Or is this already implied by the test statistics ?