The factor you deliberately change in an experiment (e.g., fertilizer amount, drug dose). It is the presumed cause in a cause-and-effect relationship.

The outcome you measure that may change in response to the independent variable (e.g., plant height, enzyme activity, survival rate).

The smallest entity that can independently receive a treatment (e.g., a pot with one plant, a single mouse, a culture flask). It defines your true sample size.

An independent experimental unit under the same conditions, used to capture natural variation and support generalizable conclusions.

Repeated measurements on the same sample, used to estimate measurement error but not to increase the number of independent data points.

A design strategy where data collectors or analysts do not know which treatment each sample received, reducing observer bias.

What is Randomization?

The process of assigning treatments to experimental units by chance, helping to balance unknown factors across groups.

What is Power analysis?

A statistical method used before data collection to estimate the sample size needed to detect a specified effect size with a chosen level of confidence.

What is Negative control?

A control group that should not show the experimental effect, used to detect background signals or contamination.

Designing Strong Biological Experiments — BIO 204: Fundamentals of Scientific Inquiry in the Biological Sciences

From Question to Experimental Blueprint

Sharpen the Question

Move from vague to testable. Instead of "Does fertilizer affect plant growth?", ask: "Does adding 10 g of nitrogen fertilizer per week increase Arabidopsis height after 4 weeks compared with no fertilizer?"

Form a Hypothesis

A hypothesis predicts an outcome and gives a reason: nitrogen-fertilized plants will be taller after 4 weeks than unfertilized plants because nitrogen supports biomass production.

Identify Variables

Independent variable: fertilizer amount. Dependent variable: plant height. Controlled variables: light, temperature, soil type, water volume, plant age, genotype.

Experimental Unit

The experimental unit is the smallest entity that independently receives a treatment, such as a pot of plants or an individual mouse, not each leaf or cell.

Treatments, Controls, and Comparison Groups

Treatment Groups

Treatment groups receive the condition of interest, such as plants getting 10 g/week nitrogen fertilizer. They are what you are testing.

Control Groups

Controls are comparison groups. Negative controls should show no effect, while positive controls confirm your system and measurements are working.

Experimental vs Observational

Experimental studies assign treatments; observational studies only measure existing differences. Experimental designs support stronger causal conclusions.

Confounding Variables

A confounder varies with the treatment and affects the outcome, like fertilized plants all on a sunny bench. Mix and randomize to avoid this.

Worked Example: Designing a Plant Growth Experiment

Question and Hypothesis

Question: Does nitrogen fertilizer increase Arabidopsis height after 4 weeks? Hypothesis: plants with 10 g/week nitrogen will be taller than unfertilized plants.

Variables and Units

Independent: fertilizer level. Dependent: height after 4 weeks. Experimental unit: one plant per pot. Control soil, pot size, light, water, and age.

Groups and Controls

Groups: negative control (0 g/week), treatment (10 g/week). Optionally add 5 g/week to explore a dose-response pattern.

Design Steps

Randomly assign pots to groups, grow on the same shelf, shuffle positions weekly, water equally, fertilize only treatments, then measure height at 4 weeks.

Replication, Pseudoreplication, and Sample Size

Biological vs Technical Replicates

Biological replicates are independent units (different plants, animals). Technical replicates are repeated measurements on the same sample and do not increase n.

Pseudoreplication

Pseudoreplication occurs when non-independent measurements, like 10 leaves from one plant, are treated as independent. Here, the true n is 1 plant.

How Many Replicates?

More biological replicates improve precision and power. Very small n, like 3 per group, often cannot detect moderate effects reliably.

Practical Focus

Aim for several biological replicates per group, define your experimental unit clearly, and remember that more replicates reduce the impact of random noise.

Randomization and Blinding to Reduce Bias

Randomization

Assign treatments by chance, not by choice. Use a random number generator to decide which pots get fertilizer and where they sit on the shelf.

Why Randomize?

Randomization spreads unknown influences, like subtle light or temperature differences, across all groups, reducing systematic bias.

Blinding

Blinding means the measurer does not know which sample is which group. Use coded labels so expectations cannot influence measurements.

Other Bias Sources

Watch for selection bias, measurement bias, and survivorship bias. Replace subjective choices with randomization and clear rules.

Design It Yourself: Improve a Flawed Experiment

Read this flawed experiment, then answer the prompts (mentally or in notes).

Scenario

A student wants to test whether a new nutrient solution speeds up yeast growth. They do this:

Prepare one flask with standard medium (control) and one with nutrient solution (treatment).
Inoculate both with yeast from the same starter culture.
Place the control flask on a shaker at 30°C, and the treatment flask on a bench at room temperature.
After 24 hours, measure optical density (OD) once for each flask.

Your tasks

Identify at least three problems with this design. Think about replication, controls, confounders, and bias.
Redesign the experiment to fix them.

Use these guiding questions:

What is the experimental unit here? Is there true replication?
Are there any confounding variables between the two flasks?
How could you add biological replicates?
How would you randomize or blind, if possible?

Pause and sketch your improved design in a few bullet points before moving on.

Check Your Understanding: Replication and Bias

Answer this question to test your understanding.

In a study on drug effects in mice, a researcher has 20 mice. They put all 10 drug-treated mice in the top cage of a rack and all 10 control mice in the bottom cage. They measure weight gain after 2 weeks. Which statement best describes the main design problem?

There is no replication; the sample size is effectively 1 per group.
There is a confounding variable because cage position is linked to treatment.
Technical replicates should have been used instead of biological replicates.
The independent and dependent variables are reversed.

Show Answer

Answer: B) There is a confounding variable because cage position is linked to treatment.

All mice in each group share the same cage position, so cage position (top vs bottom) is perfectly confounded with treatment. If temperature, light, or disturbance differ by cage position, you cannot tell whether weight changes are due to the drug or the cage. There is replication (10 mice per group), and the issue is not about technical replicates or variable naming.

Using Simple Code to Randomize Assignments

You can use simple code to randomize experimental units into groups, which helps avoid selection bias.

Here is a short Python example (you can run this in any basic Python environment, such as a Jupyter notebook or an online interpreter):

```python

import random

Example: randomly assign 12 plant pots to control or fertilizer

total_pots = list(range(1, 13)) # pots labeled 1 to 12

Shuffle the list in place for random order

random.shuffle(total_pots)

Split into two equal groups (6 control, 6 treatment)

controlpots = totalpots[:6]

fertilizerpots = totalpots[6:]

print("Control pots:", control_pots)

print("Fertilizer pots:", fertilizer_pots)

```

How this helps your experimental design

Ensures that you do not (even unconsciously) choose the “best-looking” pots for a particular group.
Makes your assignment reproducible if you save the random seed (e.g., `random.seed(42)` before shuffling).
Encourages transparent, pre-planned design rather than adjusting groups after seeing early results.

Key Terms Review

Flip through these cards to reinforce the core concepts.

Independent variable: The factor you deliberately change in an experiment (e.g., fertilizer amount, drug dose). It is the presumed cause in a cause-and-effect relationship.
Dependent variable: The outcome you measure that may change in response to the independent variable (e.g., plant height, enzyme activity, survival rate).
Experimental unit: The smallest entity that can independently receive a treatment (e.g., a pot with one plant, a single mouse, a culture flask). It defines your true sample size.
Biological replicate: An independent experimental unit under the same conditions, used to capture natural variation and support generalizable conclusions.
Technical replicate: Repeated measurements on the same sample, used to estimate measurement error but not to increase the number of independent data points.
Pseudoreplication: Treating non-independent measurements (e.g., multiple leaves from one plant) as if they were independent replicates, which inflates sample size and misleads analysis.
Randomization: Assigning treatments to experimental units by chance to evenly distribute unknown factors and reduce systematic bias.
Blinding: Keeping the person collecting or analyzing data unaware of which samples belong to which group, to prevent expectations from influencing measurements.
Negative control: A group expected to show no effect, used to reveal background noise or contamination and to interpret whether the treatment truly caused a change.
Positive control: A group expected to show a known effect, used to confirm that the experimental system and measurement methods are functioning correctly.

Key Terms

Blinding: A design strategy where data collectors or analysts do not know which treatment each sample received, reducing observer bias.
Randomization: The process of assigning treatments to experimental units by chance, helping to balance unknown factors across groups.
Power analysis: A statistical method used before data collection to estimate the sample size needed to detect a specified effect size with a chosen level of confidence.
Negative control: A control group that should not show the experimental effect, used to detect background signals or contamination.
Positive control: A control group expected to show a known response, verifying that the experimental system and measurements are working as intended.
Experimental unit: The smallest unit that can independently receive a treatment; it determines the true sample size in an experiment.
Pseudoreplication: An error where non-independent observations are treated as independent replicates, leading to incorrect statistical conclusions.
Dependent variable: The measured outcome that is expected to change when the independent variable is altered.
Technical replicate: A repeated measurement or assay on the same sample, used to assess measurement precision rather than biological variation.
Biological replicate: An independent instance of an experimental unit under the same conditions, used to estimate natural variability.
Confounding variable: A factor that varies with the independent variable and also affects the dependent variable, making it hard to identify the true cause of an effect.
Independent variable: The factor that is deliberately changed or manipulated in an experiment to test its effect on the dependent variable.