Average Variance Extracted (AVE)
When a researcher finds the average variance extracted for a construct, he is interested in knowing, on average, how much variations in his items can be explained by the construct or latent variable.
For instance: in the table below, perceived quality of information in Wikipedia (Qu) was measured with four items and the AVE for these items is 0.658, which means, on average, 65.8% of the variations in university faculty’s perceived quality of information in Wikipedia is explained by these four items or questions. On the other hand, we recall a 34.2% error in our measurement items which is fairly okay.
| Construct | Average Variance Extracted |
|---|---|
| Quality (Qu) | 0.658 |
| Use Behavior (Use) | 0.648 |
| Incentive (Inc) | 0.668 |
| Sharing Attitude (SA) | 0.745 |
| Perceived Ease of Use (PEU) | 0.512 |
As a rule of thumb and for adequate convergent, an AVE of at least 0.50 is highly recommended. That been said, an AVE less than 0.50 means your items explain more errors than the variance in your constructs. For any measurement model, an AVE must be calculated for each construct and must be at least 0.50.