AP Statistics (AP Stats): Experimental Design Principles (Randomization, Replication, Control)

AP Statistics – Experimental Design Principles (Randomization, Replication, Control)

AP Statistics: Experimental Design Principles (Randomization, Replication, Control) – Exam-Ready Study Guide

What This Is

Experimental design principles ensure that studies produce valid, reliable results by minimizing bias and confounding variables. On the AP exam, you’ll need to design experiments, identify flaws in existing designs, and justify why randomization, replication, and control are essential. For example, if a pharmaceutical company tests a new drug, proper experimental design ensures that any observed effect is due to the drug—not outside factors like age, diet, or pre-existing conditions.

Key Terms & Formulas

• Experiment vs. Observational Study:
• Experiment: Researchers actively impose treatments to measure responses (e.g., giving a drug to some patients and a placebo to others).

• Observational Study: Researchers observe and record data without intervening (e.g., surveying people about their exercise habits and heart health).

• Treatment: A specific condition applied to experimental units (e.g., "50mg of Drug X" or "30 minutes of exercise daily").

• Experimental Units: The smallest collection of individuals to which treatments are applied (e.g., patients, plots of land, lab mice).

• Response Variable: The outcome measured (e.g., blood pressure, crop yield, test scores).

• Explanatory Variable (Factor): The variable manipulated to see its effect (e.g., drug dosage, fertilizer type).

• Randomization: Randomly assigning experimental units to treatments to balance confounding variables (e.g., flipping a coin to decide who gets the drug vs. placebo).

• Replication: Using enough experimental units in each treatment group to reduce variability and detect true effects (e.g., testing the drug on 100 people, not 5).

• Control: Keeping other variables constant to isolate the effect of the treatment (e.g., giving all patients the same diet, ensuring they take pills at the same time).

• Control Group: A baseline group that receives no treatment or a placebo (e.g., patients given a sugar pill instead of the drug).

• Placebo Effect: When subjects respond to a fake treatment because they believe it works (e.g., patients reporting less pain after taking a sugar pill).

• Blinding:

• Single-blind: Subjects don’t know which treatment they receive (reduces placebo effect).

• Double-blind: Neither subjects nor researchers know who gets which treatment (reduces bias in measurement).

• Blocking: Grouping experimental units by a similar characteristic (e.g., age, gender) before randomizing treatments to reduce variability (e.g., testing a drug separately on men and women).

• Matched Pairs Design: A special case of blocking where pairs of similar units are matched, and one gets Treatment A, the other Treatment B (e.g., testing two sunscreens on the same person’s left and right arms).

Step-by-Step / Process Flow

How to Design an Experiment (AP FRQ Style)

1. State the Goal:
2. Clearly define the response variable and explanatory variable(s).

3. Example: "We want to test if a new fertilizer increases tomato yield (response) compared to the standard fertilizer (explanatory)."

4. Identify Experimental Units:

5. Determine who/what will receive treatments (e.g., 60 tomato plants).

6. Randomize Assignment:

7. Use a random method (e.g., randInt(1,60) on TI-84) to assign units to treatment groups.

8. Why? Balances confounding variables (e.g., sunlight, soil quality).

9. Apply Treatments & Control Extraneous Variables:

10. Ensure all groups are treated identically except for the explanatory variable (e.g., same water, sunlight, pot size).

11. Include a control group if needed (e.g., plants with no fertilizer).

12. Replicate:

13. Use enough units per group (e.g., 30 plants per fertilizer type) to detect true effects.

14. Measure & Compare:

15. Record the response variable (e.g., tomato yield in grams) and compare groups using statistical tests (e.g., two-sample t-test).

16. Justify Your Design:

17. Explain how randomization, replication, and control reduce bias and variability.

Common Mistakes

• Mistake: Forgetting to randomize and instead assigning treatments based on convenience (e.g., giving the new drug to the first 20 patients who arrive).

• Correction: Always randomly assign treatments (e.g., using randInt or a random number table). Why? Non-random assignment can introduce bias (e.g., early patients may be healthier).

• Mistake: Confusing observational studies with experiments.

• Correction: Experiments impose treatments; observational studies do not. Why? Only experiments can establish causation (e.g., "Drug X causes lower blood pressure").

• Mistake: Ignoring replication (e.g., testing a drug on only 3 people).

• Correction: Use large enough sample sizes to detect true effects. Why? Small samples lead to high variability and unreliable results.

• Mistake: Not controlling extraneous variables (e.g., testing a weight-loss drug while letting subjects eat whatever they want).

• Correction: Keep all other variables constant (e.g., same diet, exercise plan). Why? Confounding variables can mask or exaggerate the treatment effect.

• Mistake: Forgetting blinding when it’s possible.

• Correction: Use single- or double-blinding to reduce bias. Why? The placebo effect can distort results (e.g., patients reporting improvement just because they think they got the drug).

AP Exam Insights

• FRQ Hot Topics:
• Designing an experiment (e.g., "Describe how to test if a new teaching method improves test scores").
• Identifying flaws in a given experiment (e.g., "Explain why this study’s results may be biased").

• Justifying randomization/replication/control (e.g., "Why is blocking by gender important in this study?").

• Tricky Distinctions:

• Random Sampling (for surveys) vs. Random Assignment (for experiments):
  • Random sampling-Generalizes to a population.
  • Random assignment-Establishes causation in an experiment.

• Blocking vs. Stratifying:

  • Blocking-Used in experiments to reduce variability (e.g., testing a drug separately on men and women).
  • Stratifying-Used in sampling to ensure representation (e.g., surveying equal numbers of men and women).

• Calculator Pitfalls:

• Don’t use rand for random assignment—use randInt (e.g., randInt(1,50) to assign 50 plants to 2 groups).
• For matched pairs, remember to use paired t-tests (not two-sample t-tests).

Quick Check Questions

1. (Multiple Choice)

A researcher wants to test if a new energy drink improves reaction time. She recruits 50 volunteers, gives the drink to the first 25 who arrive, and measures their reaction times. The last 25 serve as a control group. What is the biggest flaw in this design? - (A) Lack of replication - (B) No control group - (C) Lack of randomization - (D) No blinding

Answer: (C) Lack of randomization. Why? The first 25 volunteers may differ systematically from the last 25 (e.g., more motivated, earlier risers).

2. (FRQ Part)

A farmer wants to test if a new fertilizer increases corn yield. He has 40 plots of land and plans to apply the new fertilizer to 20 randomly selected plots, leaving the other 20 as a control group. - (a) Explain why randomization is important in this experiment. - (b) Suppose the farmer also wants to account for differences in sunlight exposure. Describe how he could use blocking in this experiment.

Answer: - (a) Randomization balances confounding variables (e.g., soil quality, water access) between the two groups, ensuring that any difference in yield is due to the fertilizer. - (b) The farmer could block by sunlight exposure (e.g., group plots into "high sun" and "low sun" blocks), then randomly assign half of each block to the new fertilizer and half to the control. This reduces variability due to sunlight.

Last-Minute Cram Sheet

1. Experiments vs. Observational Studies: Experiments impose treatments; observational studies do not.
2. 3 Principles of Experimental Design: Randomization (balance confounders), Replication (reduce variability), Control (isolate treatment effect).
3. Random Assignment: Use randInt(1,n) on TI-84 to assign treatments.
4. Control Group: A baseline group (no treatment or placebo).
5. Blinding: Single-blind (subjects don’t know), double-blind (neither subjects nor researchers know).
6. Blocking: Group similar units before randomizing (e.g., by age, gender).
7. Matched Pairs: Compare two treatments on the same unit (e.g., left vs. right arm).
8. Placebo Effect: Subjects respond to a fake treatment.
9. Always justify why randomization/replication/control are used in FRQs!
10. Experiments can establish causation; observational studies cannot.