The Data Dog Credo
Gather diverse real-time evidence before acting under uncertainty
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 99%
The Data Dog Credo is a decision process for uncertain environments: distinguish the time for walking, when evidence supports action, from the time for sniffing, when clarity is insufficient. During sniffing, the decision-maker gathers every relevant data series across multiple frequencies and avoids allowing one familiar metric to stand in for the whole system. Quantitative indicators are triangulated with private-sector information, ratios, behavioral measures, and direct conversations with people operating in the field. The method accepts that a strong theory may occasionally reach the answer faster, but prioritizes robustness when cycles are unprecedented and historical analogies are unreliable. Its output is a conditional decision supported by converging evidence and explicit readiness to revise when the data change.
Origin
Austan Goolsbee described himself as a data dog rather than a monetary-policy hawk or dove, using a kennel metaphor to explain his evidence-first approach.
Core principles
- 01Know when evidence supports action and when uncertainty demands investigation
- 02Use multiple measures for every important question
- 03Combine quantitative data with direct field intelligence
- 04Prefer current, diverse evidence in unprecedented conditions
- 05Remain willing to change the view when the evidence changes
How to run it
- 1
Choose between walking and sniffing
Assess whether the situation is clear enough for action or whether material uncertainty remains. Move into investigation mode when ambiguity could change the decision.
Pro tip Name the missing evidence that would alter your current choice.
Watch out Perpetual sniffing can become analysis paralysis when evidence is already sufficient.
- 2
Gather broadly
Collect relevant data at different frequencies and from both official and private sources. Do not discard an indicator merely because it is unconventional.
Pro tip Include high-frequency measures that may reveal turning points before standard reports.
Watch out More data are useful only when they bear on the decision.
- 3
Triangulate each system
Use several measures rather than relying on one headline statistic. For labor markets, for example, combine employment, unemployment, participation, vacancies, hiring, and quits.
Pro tip Use ratios when population changes distort absolute levels.
Watch out Correlated measures can create false confidence while repeating the same underlying signal.
- 4
Add field intelligence
Talk directly with businesses and other participants close to current conditions. Use their observations to test and interpret the formal data.
Pro tip Ask operators what changed recently and what they expect to change next.
Watch out Anecdotes should complement broad evidence, not override it.
- 5
Find the coherent signal
Look for convergence across independent sources and identify the most defensible through-line. State remaining uncertainty and the evidence that would reverse the conclusion.
Pro tip Record competing explanations before selecting one.
Watch out Do not force conflicting evidence into a prematurely simple story.
- 6
Walk and keep monitoring
Act once the evidence is coherent enough, then continue watching the indicators most likely to disprove the decision. Revise without attachment when conditions change.
Pro tip Precommit to review triggers before implementing the decision.
Watch out Calling yourself data-driven is meaningless if contrary data cannot change your view.
In the wild
Payroll growth becomes difficult to interpret because immigration and population estimates are changing. Instead of relying on payrolls alone, an analyst examines unemployment, labor-force participation, vacancies, unemployment-to-vacancy ratios, hiring, quits, and reports from regional businesses.
→ The decision rests on a multidimensional assessment rather than a distorted headline statistic.
A founder sees conflicting conversion data before a launch. She reviews daily and weekly cohorts, interviews recent users, compares activation with retention, and sets explicit thresholds for proceeding or delaying.
→ The launch decision reflects converging behavioral and qualitative evidence while preserving clear revision triggers.
Common mistakes
Following one headline metric
A single indicator may be distorted by population shifts, participation changes, measurement lags, or unusual conditions.
Collecting without deciding
The point of sniffing is to resolve consequential uncertainty, not postpone action indefinitely.
Using anecdotes as proof
Field conversations add timeliness and context but cannot substitute for representative evidence.
Is it for you?
Best for
It is best for leaders and analysts making consequential decisions in uncertain or unprecedented environments.
Not ideal for
It is not ideal when a fast, reversible decision has little downside and further research would cost more than an error.
From the transcript
“The first rule of the data dog kennel is that there's a time for walking, and there's a time for sniffin, and know the difference”
“go get every data series you can, every frequency, don't throw anything away.”
“If you have a question, get out there and sniff.”
From the episode
At the Money: Chicago Fed President Austan Goolsbee on Tariffs, Supply Chains and Inflation
Austan Goolsbee