Control Experiments

Part of Germ Theory

The logic and practice of using control groups to distinguish causation from coincidence in medical and scientific experiments.

Why This Matters

Without control experiments, it is impossible to know whether a treatment actually works or whether patients recovered on their own. Throughout most of human history, healers credited their treatments with cures that would have happened anyway — and continued using harmful treatments because some patients survived despite them. Bloodletting persisted for centuries partly because of this reasoning: the patient was bled, the patient recovered, therefore bleeding helped.

The control experiment is the tool that allows you to see clearly. By comparing an outcome in a treated group against the same outcome in an untreated group — all other factors equal — you can determine whether the treatment actually changed anything. This is not mere academic interest. In a medical crisis, using an ineffective treatment means using resources that could go to proven interventions, and it may actively harm patients.

In a rebuilding context, the ability to run simple comparative experiments — comparing boiled water to unboiled water, wound care with honey versus without, herb A versus herb B for fever — allows a community to build a genuinely evidence-based local medicine without access to any outside knowledge.

The Logic of Controls

Consider this scenario: you treat 10 patients with pneumonia using willow bark tea. Eight recover. Is willow bark tea an effective treatment for pneumonia?

You cannot know without a control group. Pneumonia has a natural recovery rate — many patients recover without treatment, particularly younger and otherwise healthy individuals. If the background recovery rate is also 80%, willow bark tea contributed nothing. If the background rate is 40%, it appears to have helped. If the background rate is 95%, your treatment may actually have been harmful.

A control group is a set of subjects who are identical to the experimental group in all relevant ways (age, disease severity, nutrition, living conditions) but who do not receive the treatment being tested. By comparing outcomes between groups, the effect of the treatment is isolated from all the other factors.

Variables:

• Independent variable: What you’re testing (the treatment)
• Dependent variable: What you’re measuring (recovery, mortality, fever duration)
• Controlled variables: Everything else held constant (diet, housing, care quality, disease severity)

The more controlled variables you can actually hold constant, the more informative the experiment.

Designing a Simple Control Experiment

Step 1: Define a clear question. Not “does herb X help sick people?” but “does a daily infusion of herb X reduce fever duration in adults with febrile illness compared to equivalent rest and water intake alone?”

Step 2: Define clear outcome measures. “Feeling better” is too vague. Measurable outcomes: days until fever resolves (defined as temperature below 37.5°C for 24 hours), days until return to normal activity, mortality at 14 days.

Step 3: Assign subjects to groups. The ideal assignment is random — patients are allocated to treatment or control by chance (coin flip, drawing lots), not by the healer’s judgment or patient preference. Random assignment prevents unconscious selection bias, where you might put stronger patients in the treatment group and weaker ones in the control.

Step 4: Blind where possible. A blinded experiment is one where subjects (and ideally the observer) do not know which group they are in. This prevents the placebo effect (feeling better because you believe you’re being treated) from contaminating results. In simple field conditions, full blinding is difficult, but partial blinding (the person measuring outcomes not knowing who received treatment) helps.

Step 5: Run both groups simultaneously and under identical conditions. A study run in summer compared to a control run in winter confounds the results with seasonal factors. Run both groups at the same time.

Step 6: Record and compare results. Tally outcomes for each group. Calculate the rate difference. If the treatment group has 20% better outcomes, that is the estimated treatment effect.

Field-Practical Control Experiments

Water disinfection efficacy test:

• Group A: 20 people drink boiled water for 4 weeks
• Group B: 20 comparable people drink unboiled water from the same source
• Outcome: episodes of diarrheal illness per person-week
• This directly tests whether boiling water reduces diarrhea in your specific community, with your specific water source

Wound dressing comparison:

• 30 similar wounds: 15 dressed with honey, 15 with plain boiled cloth
• Outcome: days to healing, rate of infection (redness, pus, fever)
• Practical and directly applicable to your supplies

Herbal fever treatment:

• 20 febrile patients randomized to willow bark tea vs. plain warm water
• Outcome: hours to fever resolution
• Directly tests the local herb against the default of supportive care

Sample size considerations: Small experiments (10-20 subjects per group) can detect large treatment effects (50%+ difference) but cannot reliably detect small effects. In a medical emergency with limited subjects, a result that shows a large clear difference is more credible than a marginal one. If results are ambiguous, more subjects or better-controlled conditions are needed.

Confounding and How to Recognize It

A confounder is a variable that is associated with both the treatment and the outcome, creating a false appearance of a relationship.

Example: You observe that communities using copper water vessels have lower rates of diarrhea. You conclude copper vessels prevent diarrhea. But wealthier households are more likely to own copper vessels and also have better nutrition, sanitation, and housing. Wealth is a confounder — it is associated with both copper vessel ownership and lower diarrhea rates. To isolate the copper effect, you need to control for wealth (compare copper vs. non-copper within the same wealth level) or run a randomized trial (randomly assign copper vessels to some households).

Common confounders in medical experiments:

• Disease severity: Sicker patients have worse outcomes regardless of treatment
• Age: Older and very young patients often respond differently
• Nutrition status: Malnourished patients have impaired immune response
• Concurrent illness: Multiple conditions complicate recovery
• Care quality: Better-cared-for patients recover faster regardless of treatment

Recognizing confounded results: If your treated group was systematically different from your control group in a variable that affects outcomes, the result is confounded. Document subject characteristics and check whether groups were truly comparable.

Negative Controls and Positive Controls

Negative control: A condition you are confident will produce no effect. Used to verify that your measurement system is working correctly. Example: test your disinfection protocol on sterile water — it should show no pathogen growth. If it does, your detection method is flawed.

Positive control: A condition you are confident will produce the expected effect. Example: test your new antibiotic on a known susceptible organism. If it does not inhibit growth, your preparation is wrong. Including a positive control validates that the experiment can detect a real effect if one exists.

Running both positive and negative controls alongside experimental trials allows you to interpret results with greater confidence and catch methodological errors before they lead to wrong conclusions.

Recording and Communicating Results

Keep written records of all experiments:

• Date, location, conditions
• Number of subjects in each group, their characteristics
• Treatment given (dose, frequency, duration)
• Outcome measure and results for each group
• Your interpretation and confidence level

A community that builds a written record of its experimental results over time develops a genuine evidence base for its medical practice — a local pharmacopoeia grounded in tested observation rather than tradition or authority.