Statistical strength is a measure of result reliability. Conversion Lift Testing helps you understand the value of your ads by measuring the performance of your ad campaigns at driving conversions on LinkedIn and LinkedIn Audience Network using LinkedIn Conversions API data.
Important to know
- We’re gradually releasing Conversion Lift Testing, and you might not have access to it at this time.
- Some lift metrics may be impacted due to partial traffic measurement, as certain types of traffic such as CTV Ads, iOS conversions, and unconsented traffic are currently not measured. These limitations are due to ongoing improvements in attribution, legal requirements, and our commitment to protecting member privacy.
- Statistical strength is only available for relative conversion lift results and conclusive results aren't guaranteed.
P-value and statistical strength
P-value is an important aspect of determining statistical strength as it’s the probability that your result (or a more extreme result) was caused by random chance (subject to certain assumptions). For example, if the p-value is 0.05, then there’s a five percent probability that your result (or a more extreme result) was caused by random chance.
P-value is calculated based on the two-sample bootstrap test of the relative lift metric, which is more powerful in detecting the conversion lift than the absolute lift metric. When reviewing the relative and absolute conversion lift results of a Conversion Lift test, you can select the statistical strength rating to view the p-value.
Statistical strength ratings range from very strong to very weak. In addition to reaching a certain p-value, a statistical strength rating of very strong, strong, or medium must receive at least four conversions for each of the test and control groups. For example, if the p-value is 0.04 but only two test group conversions were received, then the statistical strength is weak.
The following table provides examples of statistical strength, p-value requirement, and conversion volume requirement. We don't recommend making decisions based on results with weak or very weak statistical strength.
| Statistical strength | P-value requirement | Conversion volume requirement |
|---|---|---|
| Very strong | <=0.05 | For each of the test and control groups: Total conversions ≥ 4 |
| Strong | <=0.10 | For each of the test and control groups: Total conversions ≥ 4 |
| Medium | <=0.20 | For each of the test and control groups: Total conversions ≥ 4 |
| Weak | <=0.30 | No requirement |
| Very weak | >0.30 | No requirement |
Conversion actions like page views and add to cart are more likely to yield statistically significant results. Therefore, we recommend measuring not only lower-funnel conversion actions, such as purchases, but also mid-funnel actions. These mid-funnel actions can provide valuable insights into how members are engaging with and responding to ads.