t-Test
Data Entry
The groups may contain different numbers of values; the columns are read independently of each other.

What it does

The t-test assesses whether the difference between the means of two groups is small enough to be explained by measurement variability or reflects a real difference. It is intended for two groups only and requires continuous data (percentage dissolved, content in mg, hardness in N, % moisture and the like).

Which test to choose?

Independent samples: two separate groups with no one-to-one pairing between them. Typical uses in pharmaceutical technology:

  • Comparing the mean active-ingredient content of two production batches
  • Comparing two formulations or two granulation methods in terms of tablet hardness, friability, weight variation or disintegration time
  • Percentage dissolved of the reference and test products at a single specified time point
  • Comparing groups produced with two different compression forces, drying temperatures or pieces of equipment

Paired samples: each row holds two measurements of the same unit; the row order must be preserved. Typical uses:

  • Assay values of the same batch at the start (t=0) and after stability storage (e.g. month 6)
  • The same samples measured by two analytical methods (e.g. HPLC and UV) in method-comparison/validation studies
  • Moisture content of the same granule lot before and after drying
  • Measurements on the same subject before and after administration
Prerequisites
  • Independence: the measurements must be independent of each other. Repeated readings of the same tablet are not independent; they are replicates of a single observation and do not count as separate data points.
  • Normality: in the independent test each group, and in the paired test the differences, are expected to be approximately normally distributed. The smaller the sample, the more critical this assumption becomes.
  • Homogeneity of variance (independent test only): this application checks it with Levene's test. If p < 0.05 it switches automatically to the Welch correction and computes the degrees of freedom accordingly; no further action is needed on your part.
  • Representative sampling: the samples must have been taken so that they represent the batch.
  • Row integrity in the paired test: a row holding only one value is excluded from the analysis; the application reports the row number.
Limitations
The absence of a significant difference does not mean "equivalent". p > 0.05 does not say that there is no difference; it says that the data at hand provide no evidence of one. In a small sample a real difference is easily missed. A claim of equivalence requires an equivalence test (e.g. TOST) or a predefined acceptance criterion.
  • Not suitable for whole dissolution profiles. Running a separate t-test at every time point inflates the false-positive rate. Profiles are compared with the f1/f2 factors — the site's f1/f2 module is there for that purpose.
  • Not for more than two groups. Successive t-tests must give way to ANOVA; otherwise the error rate accumulates with every comparison.
  • Sensitive to outliers. It is built on the mean and standard deviation; a single extreme value can decide the result. The data should be inspected visually before analysis.
  • p alone is not enough. In a large sample even practically negligible differences come out significant. The confidence interval and the Cohen's d effect size given in the results panel should therefore be interpreted together: a difference may be statistically significant yet pharmaceutically unimportant.
  • It does not replace tests with a pharmacopoeial criterion. For matters such as content uniformity the acceptance criterion of the relevant pharmacopoeia applies; the t-test cannot be used in its place.
  • This application computes a two-tailed p-value: it answers "is there a difference?", not "is A greater than B?".
Test Parameters
An error has occurred. This application may no longer respond until reloaded. Reload 🗙