qPCR replicates and statistics: from counting n to error bars

About 7 min read · Updated 2026-09-30

Before you put a p-value on qPCR data, check two things: is n an independent experimental unit, and on which scale did you run the statistics?

Try it with the tool

Load several files and each plate is normalized to its own control, then combined with Welch’s t-test and error bars.

Open the qPCR tool →

1. Technical vs biological replicates

Type	What it is	What it captures
Technical	The same cDNA in several wells	Pipetting and instrument noise; average to one value per sample
Biological	Independently grown cells, different animals, different days	Real biological variation — the n for statistics

RNA from one dish measured in three wells is n = 1, not n = 3. Publications usually need at least three biological replicates.

In statistics, n is the number of biological replicates (3 here). Average each sample's three wells (technical replicates) into one value first. Treating 9 wells as n = 9 makes the p-value misleadingly small.

2. Test on the log scale

Fold change is asymmetric (a 2-fold increase and a 2-fold decrease are 2 and 0.5). Run t-tests on ΔCt or log₂ fold change (= −ΔΔCt), which are closer to normal and treat increases and decreases symmetrically.

Half (0.5×) and double (2×) are changes of equal size, yet on a linear scale their distances from 1 look different (0.5 vs 1). On a log₂ scale they become −1 and +1 — symmetric, so means and tests are accurate.

3. Which test

Two groups (control vs treatment): a t-test; Welch’s t-test is a safe default when variances may differ.
Three or more groups: one-way ANOVA with a post-hoc test — Dunnett for comparisons against control only, Tukey for all pairs.
If each control is paired with a treatment from the same day, consider a paired test.

4. Error bars: SD vs SEM

SD: how spread out the values are — it shows the data’s variability and is preferred by most journals.
SEM (= SD ÷ √n): how precise the mean is — shorter bars that can understate the spread.

Either way, state it in the legend, e.g. “mean ± SD, n = 3”, and ideally plot individual values as dots.

5. Combining plates

Experiments on different days shift overall Ct slightly. Normalize each plate to its own control (= 1) before combining and most of that shift disappears — so always run the control on every plate.