Jensen's Inequality: Why the Average Isn't What You Think

Your grandmother spends one hour in a freezer at 0°F.

Then she spends one hour in a sauna at 140°F.

Her average temperature over the two hours is 70°F.

Optimal comfort temperature.

She dies.

This is Jensen's Inequality made intuitive. The problem: temperature's effect on her health is nonlinear. The danger accelerates at the extremes. An hour at 140°F is not just a mild inconvenience that balances an hour at 0°F.

The average is mathematically correct. The average of 0 and 140 is 70. But the average is useless for predicting her health outcome.

Why? Because her response to temperature is concave — the harm increases nonlinearly at the extremes.

The Mathematical Statement

Jensen's Inequality states:

For a concave function, the function of the average is greater than or equal to the average of the function.

Translated into English: for any system whose response is concave (curves inward), the average outcome is worse than the outcome of the average.

Applied:

• If you take the average temperature and then apply the system's response, you get one number
• If you apply the system's response to each temperature and then take the average, you get a worse number

The non-linearity creates a gap.

Where This Matters

Sleep:

Average of 8 hours is supposedly optimal. But 4 hours one night, 12 hours the next night (averaging 8) is not the same as consistently 8 hours.

Why? Your sleep debt is nonlinear. Missing 4 hours creates a deficit that sleeping 12 hours doesn't fully recover. The damage accumulates nonlinearly.

Exercise:

Average of 3x per week is good. But once one week, 5x the next week (averaging 3) is worse than a consistent 3x/week.

The body's adaptation to stress is nonlinear. Inconsistent stimulus produces inconsistent adaptation.

Income volatility:

Average income of $50K is stable. But $30K one year, $70K the next (averaging $50K) is worse for financial planning and wellbeing.

The stress response to income uncertainty is nonlinear. The uncertainty itself is worse than a steady income at the average.

Project timelines:

Average completion time of 3 months is fine. But 1 month then 5 months (averaging 3) is worse because you can't plan, clients are frustrated, and rework happens.

The project's success doesn't depend on the average duration but on the consistency.

The Practical Implication

For any system where the response is concave (and most living systems are concave):

The stability of input matters more than the average input.

A consistent 70°F is better than averaging 70°F through extremes.

A consistent 8 hours of sleep is better than averaging 8 hours with extremes.

A steady income of $50K is better than averaging $50K with volatility.

This is why:

• Consistency is underrated. Most advice focuses on the average (8 hours sleep, 3x/week exercise, $50K income). The advice should focus on consistency.

• Volatility is harmful in concave systems. If you're fragile (concave), volatility damages you more than the average suggests.

• Measurement of averages is misleading. When you hear "average temperature," or "average return," or "average time to completion," you should immediately ask: what's the variability?

Why This Is Hidden

Jensen's Inequality is hidden because:

1. Averages are easy to calculate and communicate. It's simple to say "average of 70°F" rather than explain nonlinearity.

2. Most statistics are presented as averages. Medical research, economic reports, business metrics — all lean on averages.

3. The gap is invisible. If you don't know the distribution, you can't see that the average is misleading.

4. Prediction is hard. If you're building a model to predict outcomes, using the average is wrong, but you won't know it until the model fails.

Real-World Examples

Traffic:

Cities report "average commute time 30 minutes." But if the distribution is 10 minutes on good days, 60 minutes on bad days, the psychological experience is worse than a consistent 30 minutes.

The inconsistency itself is a cost that the average doesn't capture.

Medication:

A drug might average to a safe dose, but the dose variability might mean that some people get too much (harmful) and some too little (ineffective), while the average is fine.

The average hides the nonlinear response at the extremes.

Stock returns:

An investment might average 10% annually, but with extreme volatility (down 50% one year, up 100% the next). The average is fine; the volatility is harmful because it hits your loss limit and forces liquidation.

The average doesn't capture the full risk.

The Solution

When you see an average:

1. Ask for the distribution. What's the range? What's the standard deviation?

2. Check whether the system is linear or nonlinear. For linear systems, averages work fine. For nonlinear systems (most real systems), averages hide the truth.

3. If nonlinear, optimize for consistency, not average. Build for steady outcomes rather than optimal-on-average outcomes that have extreme variance.

4. Account for the gap. If you know the system is concave, add buffer. Don't plan based on the average alone.

Jensen's Inequality is why grandmother dies at 70°F average. And it's why most averages are misleading.