A/B Testing Core Web Vitals Formula
Learn how the calculator compares control and variant LCP, INP, and CLS values and estimates overall Core Web Vitals improvement.
This calculation estimates whether an A/B test variant improves loading speed, interaction responsiveness, and visual stability relative to a control. It normalizes each Core Web Vital against a reference threshold, averages the three results, and measures the percentage reduction in the combined index.
- 100% Free
- No Sign-Up Required
- Private & Secure
- Mobile Friendly
Overall Core Web Vitals Improvement
Where:
First calculate a normalized, equal-weight index for the control and variant. Then find how much lower the variant index is as a percentage of the control index. A positive result favors the variant.
Variables Explained
| Variable | What It Means | Unit |
|---|---|---|
| controlLcp - Control LCP | Largest Contentful Paint for the control experience. | seconds |
| variantLcp - Variant LCP | Largest Contentful Paint for the test variant. | seconds |
| controlInp - Control INP | Interaction to Next Paint for the control experience. | milliseconds |
| variantInp - Variant INP | Interaction to Next Paint for the test variant. | milliseconds |
| controlCls - Control CLS | Cumulative Layout Shift score for the control experience. | N/A |
| variantCls - Variant CLS | Cumulative Layout Shift score for the test variant. | N/A |
| controlCwvIndex - Control CWV index | Equal-weight average of the control metrics after normalizing them against the selected reference thresholds. | number |
| variantCwvIndex - Variant CWV index | Equal-weight average of the variant metrics after normalizing them against the selected reference thresholds. | number |
Step-by-Step Calculation
Calculate LCP improvement
A positive result means the variant reaches its largest content element sooner than the control.
(controlLcp - variantLcp) / controlLcp * 100
Calculate INP improvement
A positive result means the variant has a lower interaction delay measurement.
(controlInp - variantInp) / controlInp * 100
Calculate CLS improvement
A positive result means the variant has less unexpected layout movement.
(controlCls - variantCls) / controlCls * 100
Calculate the control index
Each control metric is divided by its reference threshold and the three ratios are averaged. Lower is better.
((controlLcp / 2.5) + (controlInp / 200) + (controlCls / 0.1)) / 3
Calculate the variant index
The same equal-weight normalization is applied to the variant.
((variantLcp / 2.5) + (variantInp / 200) + (variantCls / 0.1)) / 3
Calculate overall improvement
The result shows the percentage reduction in the combined index from control to variant.
(controlCwvIndex - variantCwvIndex) / controlCwvIndex * 100
Example: Comparing a faster test variant
LCP improvement
(2.8 - 2.4) / 2.8 × 100
14.3%
INP improvement
(240 - 180) / 240 × 100
25.0%
CLS improvement
(0.12 - 0.08) / 0.12 × 100
33.3%
Control CWV index
((2.8 / 2.5) + (240 / 200) + (0.12 / 0.1)) / 3
1.17
Variant CWV index
((2.4 / 2.5) + (180 / 200) + (0.08 / 0.1)) / 3
0.89
Overall CWV improvement
(1.17 - 0.89) / 1.17 × 100
24.4%
Final Result
The variant shows an estimated 24.4% overall Core Web Vitals improvement versus the control.
Assumptions
- ✓Control and variant values use the same measurement method, date range, device mix, location mix, and traffic sources.
- ✓Lower LCP, INP, and CLS values represent better user experience outcomes.
- ✓The index gives LCP, INP, and CLS equal weight after normalizing against 2.5 seconds, 200 ms, and 0.1 respectively.
- ✓The inputs are representative summary metrics for each A/B test experience.
Limitations
- !The combined index is a simplified comparison tool, not an official ranking score or pass/fail assessment.
- !A lower index does not establish statistical significance for the observed differences.
- !Real-user metrics can change with network conditions, browser mix, page type, traffic composition, and experiment implementation.
- !Performance improvement alone does not show whether a variant improves conversion, revenue, engagement, or other business outcomes.
Common Mistakes to Avoid
Comparing field data for one version with lab data for the other.
Entering LCP in milliseconds even though the calculator expects seconds.
Treating a positive percentage as worse; positive improvement means the variant value is lower.
Comparing periods with substantially different device, geography, or traffic-source mixes.
Using only the combined index and overlooking a regression in an individual Core Web Vital.
Related Formulas
Frequently Asked Questions
How is overall Core Web Vitals improvement calculated?
The calculator averages normalized LCP, INP, and CLS ratios for each version, then calculates the percentage reduction from the control index to the variant index.
What does a CWV index of 1.00 mean?
It means the average of the three normalized metrics equals the reference thresholds used by the calculator: 2.5 seconds LCP, 200 ms INP, and 0.1 CLS.
Is a lower Core Web Vitals index better?
Yes. Lower values indicate that the normalized metrics are lower relative to the calculator's reference thresholds.
Why are LCP, INP, and CLS divided by different values?
They use different units and scales. Dividing each metric by a reference threshold puts them on a comparable relative scale before averaging.
Can one metric improve while the overall result worsens?
Yes. A gain in one metric can be outweighed by regressions in the other metrics, depending on their normalized changes.
Ready to calculate your result?
Use the calculator to get instant results with your own inputs.