Two Decimal Places Is Not the Same as Two Digits of Proof
A lift reported as 37.42 percent looks more rigorous than 37 percent, but trailing decimals rarely survive the sample size behind them. Here is how to read them.
A lift reported as 37.42 percent looks more rigorous than 37 percent, but trailing decimals rarely survive the sample size behind them. Here is how to read them.
A case study can name its sample size and still hide the response rate, and a low rate quietly turns a big sample into a self-selected one.
An indexed result like a consideration index of 140 is a percentage in disguise. Here is why you refuse to read one until the base period is named.
An incrementality result only holds if the control was genuinely unexposed. Here is how shares, spillover, and shared devices quietly contaminate the holdout.
Trimming outliers before averaging is legitimate, but an undisclosed trim hands you a cleaner number than the data holds. Here is how to ask which average you got.
A case study’s comments and shares can hold public pushback you never went looking for. Here is how to read that layer honestly.
A campaign that ran on three platforms and reports results for one is not lying. It is silent, and silence is the part worth checking.
Some case studies compress twelve months of uneven results into one narrative. Here is how to tell, and why the unit of analysis changes the number.
A case study can outlive the exact feature, format, or tool it describes. Here is how to check before you borrow its lesson for a live campaign.
The metric name matches but the figure does not. Here is why that mismatch is the fastest check in reading a case study honestly.