2 by 2 tables ✅ check. But what if your placebo group shows no response at all? Today’s post explores how low counts and zeros challenge standard analysis and what to do about it.

 

Challenges

⚠️ When the expected number of observations is small, we can't assume a Χ² distribution.

⚠️ When computing effect measures, like the odds ratio or the risk ratio, we have to divide by entries from our 2 by 2 table.

But of course this is not possible if the divisor is zero.

 

Statistical Methods

📊 Zero-Cell-Correction: In the case of entries in the 2 by 2 table being zero, we can add a small number (e.g. 0.5) to all entries and use these new numbers for our computations. This solves the issue of dividing by zero and - since we only marginally change the numbers – doesn't skew the results. But be careful: you should only do a zero-cell-correction if only one entry is zero. If several entries are zero, the results are not meaningful.

📊 Alternative effect measure – If we have very rare events, we can use the Peto-Odds-Ratio. In this case no zero-cell-correction is necessary.

 

Take Home Messages

⚠️ 2 by 2 tables can be deceptive – small numbers or zeros can throw things off.
💪  But with the right tools, you're ready to handle them like a pro.

 

📢 Stay tuned and continue Building Statistical Confidence For Market Access