T.TEST in Collabora Office: Compare Samples with an Explicit t-Test Design

=T.TEST(Data1; Data2; Mode; Type)

Data1 and Data2 are the two numeric data ranges. Mode=1 requests the one-tail result; Mode=2 requests two tails. Type=1 is paired, Type=2 independent samples with equal variance, and Type=3 independent samples with unequal variance. All four arguments are required.

These Type settings describe different sampling designs and variance assumptions. Choosing the smallest resulting probability is not a legitimate way to select the test. Pairing requires meaningful row-by-row matches, not merely equally long lists.

The dedicated Collabora documentation identifies the modes and types. Current source uses an absolute t statistic before applying its tail calculation. Consequently Mode=1 returns the smaller magnitude-based tail; swapping the two samples does not encode a chosen directional alternative.

Use the five paired rows from the quick answer. In C2 enter =A2-B2 and fill down. Differences are 2, 4, 1, 1 and 1. Their mean is 1.8, sample variance 1.7, and sample standard deviation approximately 1.303840481.

The independent statistic is 1.8/SQRT(1.7/5), approximately 3.086974533, with 4 degrees of freedom. Applying the two-tail t probability gives approximately 0.036681989.

The paired calculation uses variation in the differences. It does not pool the two original columns’ variances as though the observations were independent.

For the same numerical columns considered as an alternative independent-sample illustration, means are 12.8 and 11, and sample variances 3.7 and 2.5.

=T.TEST(A2:A6;B2:B6;2;2) has independent equal-variance target approximately 0.144659850. Its statistic is approximately 1.616447718 with eight degrees of freedom.

=T.TEST(A2:A6;B2:B6;2;3) has independent Welch target approximately 0.146072493. Here the statistic happens to be the same because the sample sizes are equal, but Welch’s effective degrees are approximately 7.711133400. The differing probabilities illustrate the designs; they do not justify reclassifying genuinely paired data.

For the paired dataset, Mode=1 has independent target approximately 0.018340995, half of the Mode=2 result. Reversing A and B changes the mean difference’s sign but leaves this magnitude-based output unchanged in the corroborating source.

A prespecified directional alternative must account for whether the observed effect is in that direction. Do not label the smaller tail a directional p-value without checking the sign and intended hypothesis. Keep the signed mean difference and the sample ordering visible beside the probability.

For Type=1, the current source requires the two matrices to have the same dimensions and iterates numeric pairs. A missing observation should not be repaired by shifting one column upward independently, because that creates different pairings.

For Types 2 and 3, the helper skips string/empty entries separately and requires at least two usable numeric observations in each sample. Use complete numeric ranges in reproducible examples, and audit missing-data handling before analyzing real observations. These source paths do not certify every error-cell or Boolean conversion.

Mode and Type are approximately floored in the source; explicitly supply integer options rather than relying on this conversion. Unsupported options follow an illegal-argument path. A paired sample with zero variance in differences reaches a division-by-zero path; a probability cannot be recovered merely by replacing that error with zero.

T.TEST returns a probability, not an effect size, practical-importance measure or proof of a hypothesis. Small samples, dependence, outliers and the relevant distributional assumptions require separate evaluation. No native test or complete error-display inventory is claimed.

The exact function is documented in Collabora Office 26.04 desktop help. That documentation snapshot was retrieved on 2026-10-13 and hash-verified on October 15, 2026. It does not establish Online or mobile support. First introduction and last functioning release are unknown in this evidence.

Additional implementation statements come from Collabora’s current engine source, retrieved October 15, 2026. This moving-main source is corroboration, not a pinned desktop 26.04 binary or a native execution. No native Collabora Office run occurred. Worked outputs are independent mathematical expectations.

Formulas use English names and semicolon separators. The help identifies namespace COM.MICROSOFT.T.TEST, outside its cited OpenFormula 1.3 standard; that does not establish identical behavior in another application.

• Collabora Office 26.04: T.TEST.
• Collabora engine implementation.
• Editorial testing standards.