Update Often, in Small Steps

When you order food delivery, the app's estimated arrival time keeps changing: 35 minutes when you order, 42 when the restaurant is slow, back to 38 when the rider takes a shortcut. Nobody calls the app "wishy-washy." If anything, that's why you trust it.

Forecasts work the same way. Writing down a probability is the start, not the finish.

Why small, frequent steps

Researchers who studied long-running forecasting tournaments found that the best forecasters shared a habit: they updated often, but by small amounts. In a four-year study of more than 400,000 forecasts, top “superforecasters” submitted about five forecasts per question, more than twice as many as typical participants, and moved about 11 points per update, roughly half as much.

• Most new information is only "somewhat relevant." Using the method from lesson 4.1, its likelihood ratio is often between 1.2 and 2, which only justifies a small move.
• Many small updates let weak evidence add up, which protects against underreaction.
• Moving a little at a time also keeps any single headline from carrying you away, which protects against overreaction.

When a big step is fine

When decisive evidence arrives, make a big move. If you're forecasting "the company will launch its new product this month" and the official launch-event invitations have gone out, go straight to 90% or above. Small steps are the norm, not a rule you can't break.

Time itself is evidence

For forecasts of the form "X will happen by a deadline," the probability should usually drift down as the deadline approaches and nothing has happened. If you gave "launches this month" 60% on the 1st and there's still no sign of it on the 25th, staying at 60% doesn't make sense.

Write one line of reasoning for each update

When you add a probability update in FoMind, jot down a sentence on why. At review time you'll see which information actually helped and which was noise, and whether you tend to move too much or too little.