- Joined
- Jun 19, 2018
- Messages
- 6,652
- Likes
- 9,410
This was actually done in the study. The regression analysis was performed by testing the same budget bookshelf speakers consumer reports had tested. Once developed, they then applied it to larger set of speakers. That is when the correlation factor dropped from 0.99 to 0.86. It was then hypothesized that there were larger speakers in there that were not modelled as well as the smaller ones.
This is not quite right IIUC.
In Olive's second study (involving the larger sample set), he states the following:
Our anechoic model described in equation 9 [i.e. the model based on the first, smaller sample set] was applied to the new larger loudspeaker sample and produced a correlation of 0.70 between the predicted and measured preference ratings. The lower correlation was likely related to the model being too tightly fitted to the small sample (13 loudspeakers) and/or the loss of precision from combining subjective data from 18 unrelated tests. A more generalized model was necessary to accurately predict the ratings of our 70-loudspeaker sample. Using 23 independent variables, a model using 4 independent variables was developed that has a correlation of 0.86 for the 70-loudspeaker sample.
In other words, the model derived from the first (smaller) sample set produced a correlation factor of 0.70 when applied to the second (larger) sample set.
A new model was then derived from the second sample set. It was this new model that produced a correlation factor of 0.86 when applied to the same sample set from which it was derived.
Last edited: