What is a full factorial design?
A full factorial design is an experiment in which every possible combination of the selected factor levels is tested. When each of k factors is studied at two levels (a low and a high setting), the result is a 2^k design: 2² = 4 runs for two factors, 2³ = 8 runs for three, 2⁴ = 16 for four, and so on. Because no combination is skipped, the design captures the complete picture of how the factors and their interactions drive the response.
This completeness is what makes the full factorial the gold standard of design of experiments. Unlike one-factor-at-a-time testing, it varies all factors simultaneously, so it spends every run learning about multiple effects at once. And unlike fractional or screening designs, it estimates every main effect and every interaction without aliasing — nothing is confounded with anything else.
A factor is an input you control (temperature, pH, inducer concentration). A level is a setting of that factor (30°C vs 37°C). The response is the measured outcome (titer, yield, purity). A two-level 2^k design brackets each factor between a sensible low and high level and asks: across this region, which factors matter, and do they interact?
2^k run counts (and when it blows up)
The cost of a 2^k design doubles with every factor you add. That is the headline trade-off: completeness is bought with runs. Two factors cost 4 runs, but seven factors cost 128 — before you add a single replicate or center point. The table and chart below show how quickly the run count escalates.
| Factors (k) | 2^k runs | Effects estimated | Practical? |
|---|---|---|---|
| 2 | 4 | 2 main + 1 two-factor | Yes |
| 3 | 8 | 3 main + 3 two-factor + 1 three-factor | Yes |
| 4 | 16 | 4 main + 6 two-factor + higher | Yes |
| 5 | 32 | 5 main + 10 two-factor + higher | Borderline |
| 6 | 64 | 6 main + 15 two-factor + higher | Rarely |
| 7 | 128 | 7 main + 21 two-factor + higher | No — fractionate |
The practical guidance is simple. With 2–4 factors, a full factorial is affordable and gives you everything. At 5 factors (32 runs) it is borderline — justifiable only when interactions are the whole point. From 6 factors up, the run budget is usually better spent screening first with a fractional factorial design to find the critical few factors, then running a small full factorial on the survivors.
Build a 2^k design without the arithmetic
Pick your factors and levels and the free DOE generator lays out the complete coded matrix and randomizes the run order in seconds.
Main effects vs interactions
A full factorial design earns its run count by estimating two kinds of effect cleanly. A main effect is the average change in the response when a factor moves from its low to its high level. An interaction exists when the effect of one factor depends on the level of another — the reason one-factor-at-a-time experimentation so often misleads.
The arithmetic is the same for every effect and is the heart of any full factorial doe calculator. Add a column of ±1 signs for the effect, then:
Effect = (mean response where sign = +1) − (mean response where sign = −1)
For a main effect, the ±1 column is just that factor's coded level. For a two-factor interaction (say A×B), the column is the product of the two factors' signs. Because the 2^k matrix is orthogonal — every column is balanced and uncorrelated with the others — each effect is computed independently of the rest. That orthogonality is precisely what fractional designs sacrifice.
Worked 3-factor example
Here is a complete 2³ full factorial design for an E. coli soluble-protein process. Three factors are studied at two levels each: A = induction temperature (30 vs 37°C), B = IPTG (0.1 vs 1.0 mM), and C = induction OD600 (0.5 vs 1.0). The response is soluble target protein titer in mg/L. Eight runs cover every corner of the cube.
| Run | A (T) | B (IPTG) | C (OD) | Titer (mg/L) |
|---|---|---|---|---|
| 1 | − | − | − | 220 |
| 2 | + | − | − | 180 |
| 3 | − | + | − | 260 |
| 4 | + | + | − | 200 |
| 5 | − | − | + | 240 |
| 6 | + | − | + | 190 |
| 7 | − | + | + | 300 |
| 8 | + | + | + | 210 |
Computing the effects by hand
Grand mean = (220+180+260+200+240+190+300+210) / 8 = 1800 / 8 = 225 mg/L.
Main effect of A (temperature): mean at A = + (runs 2,4,6,8 = 180,200,190,210 → 195) minus mean at A = − (runs 1,3,5,7 = 220,260,240,300 → 255).
Effect A = 195 − 255 = −60 mg/L. Raising temperature from 30 to 37°C costs 60 mg/L.
Main effect of B (IPTG): mean at B = + (260,200,300,210 → 242.5) minus mean at B = − (220,180,240,190 → 207.5).
Effect B = 242.5 − 207.5 = +35 mg/L.
Main effect of C (OD): mean at C = + (240,190,300,210 → 235) minus mean at C = − (220,180,260,200 → 215).
Effect C = 235 − 215 = +20 mg/L.
A×B interaction: build the product column (sign of A × sign of B). The A×B = + runs are 1,4,5,8 (220,200,240,210 → 217.5); the A×B = − runs are 2,3,6,7 (180,260,190,300 → 232.5).
Effect A×B = 217.5 − 232.5 = −15 mg/L — the interaction the plot above shows.
Conclusion: temperature dominates, IPTG and OD help, and the negative A×B interaction means IPTG pays off most at the low temperature. The best corner is run 7 — 30°C, 1.0 mM IPTG, OD 1.0 — at 300 mg/L, well above the 225 mg/L grand mean.
Full factorial vs fractional: when to switch
A full factorial design gives you everything, but you do not always need everything. When the number of factors climbs, most of the 2^k runs are spent estimating high-order interactions (three-factor, four-factor) that are almost always negligible in biology. That is wasted effort — and the cue to switch to a fraction.
The decision comes down to your goal and your factor count:
- Optimizing 2–4 factors, interactions matter: use a full factorial design (4–16 runs). You get clean main effects and all interactions.
- Screening 5–11 factors: use a fractional factorial or Plackett-Burman design to find the critical few in a fraction of the runs, accepting that some interactions are aliased.
- Mapping curvature near an optimum: move to a response-surface design (central composite or Box-Behnken) after screening.
A common and efficient strategy is to fractionate first and finish full: screen many factors cheaply, drop the inert ones, then run a small full factorial (often augmented with center points) on the 2–3 survivors to resolve their interactions and check for curvature. The shared methodology is covered end to end in our DOE for bioprocess optimization guide.
Build it free in the DOE generator
You do not need JMP, Minitab, or a line of R to run a full factorial. A free DOE calculator generates the complete coded design matrix, randomizes the run order to guard against time trends, and lets you enter responses to compute effects — all in the browser.
Our design of experiments calculator builds 2^k full factorial designs (with optional center points and replicates), exports a clean run sheet, and works equally well as a full factorial design calculator for optimization or a screening generator for many factors. Choose your factors, set the low and high levels, and the tool returns a randomized run order you can take straight to the bench.
Generate your full factorial run sheet now
A free, no-install DOE generator: coded matrix, randomized run order, and effect analysis for your 2^k design.
Frequently Asked Questions
What is a full factorial design?
A full factorial design is an experiment that tests every possible combination of the chosen factor levels. For k factors each at 2 levels, that is a 2^k design — for example, 3 factors at 2 levels gives 2³ = 8 runs. Because every combination is run, a full factorial design estimates all main effects and all interactions without any confounding.
How many runs does a 2^k full factorial design need?
A two-level full factorial needs 2^k runs before replication: 4 runs for 2 factors, 8 for 3, 16 for 4, 32 for 5, 64 for 6, and 128 for 7. Adding center points or replicates increases this. Beyond about 5 factors the run count grows too fast and a fractional factorial or screening design is usually preferred.
What is the difference between a main effect and an interaction?
A main effect is the average change in the response when one factor moves from its low to its high level. An interaction occurs when the effect of one factor depends on the level of another — for example, if higher IPTG only helps at low temperature. Full factorial designs are the only designs that estimate every interaction cleanly, which is why they are the gold standard for studying interactions.
When should I use a full factorial design instead of a fractional one?
Use a full factorial design when you have 2 to 4 factors, you can afford 8 to 16 runs, and you need every interaction estimated without confounding — typically in the optimization phase. Switch to a fractional factorial or Plackett-Burman screening design when you have 5 or more factors and the full 2^k run count becomes impractical.
Can I build a full factorial design for free?
Yes. A free full factorial DOE calculator generates the complete coded design matrix, randomizes the run order, and analyzes main effects and interactions in your browser with no software install, no R, and no licence fee. The BioProcess Tools DOE generator builds 2^k designs and exports the run sheet directly.
Related Tools
- DOE Experiment Generator — Build full factorial, fractional factorial, and screening designs free in the browser with randomized run order.
- E. coli Expression Optimizer — Tune strain, promoter, and induction parameters — ideal factors for a 2^k design.
- Fed-Batch Feed Strategy Calculator — Generate feeding profiles to test as DOE factors.
References
- Montgomery, D.C. (2017). Design and Analysis of Experiments, 9th ed. Wiley. Chapters 6–7 (the 2^k factorial design).
- NIST/SEMATECH (2012). e-Handbook of Statistical Methods, Section 5.3.3.3: Full factorial designs. DOI: 10.18434/M32189
- Mandenius, C.F. & Brundin, A. (2008). Bioprocess optimization using design-of-experiments methodology. Biotechnology Progress, 24(6), 1191–1203. DOI: 10.1002/btpr.67