Hello,
I'm working with a dataset produced from a 5-variable full factorial screening with 3 center points. The raw data is heavily skewed with an exponential distribution. I've tried log, log10, square root and various box-cox transformations and can't seem to get anything even nearly approaching a normal distribution.
| Condition | Pattern | data1 |
| 1 | +−−+− | 0.59 |
| 2 | −+−−− | 1.6 |
| 3 | −−−+− | 1.78 |
| 4 | +−++− | 0.45 |
| 5 | +−−−− | 1.37 |
| 6 | −++−− | 0.87 |
| 7 | ++−+− | 0.05 |
| 8 | −−−−− | 4.46 |
| 9 | ++−−− | 0.14 |
| 10 | −+−+− | 0.36 |
| 11 | +++−− | 0.11 |
| 12 | +−+−− | 0.8 |
| 13 | −+++− | 0.33 |
| 14 | ++++− | 0.05 |
| 15 | −−+−− | 2.43 |
| 16 | −−++− | 1.53 |
| 17 | 0 | 0.86 |
| 18 | 0 | 0.79 |
| 19 | 0 | 0.9 |
| 20 | −+−−+ | 0.94 |
| 21 | +−+++ | 0.87 |
| 22 | −−+++ | 0.91 |
| 23 | −−−++ | 0.72 |
| 24 | −+−++ | 0.05 |
| 25 | −−+−+ | 2.74 |
| 26 | −++−+ | 0.72 |
| 27 | −++++ | 0.08 |
| 28 | +−+−+ | 2.92 |
| 29 | −−−−+ | 4.08 |
| 30 | ++−−+ | 0.88 |
| 31 | +−−−+ | 3.98 |
| 32 | +−−++ | 0.82 |
| 33 | +++++ | 0.08 |
| 34 | +++−+ | 0.78 |
| 35 | ++−++ | 0.06 |
1. What kind of transform is appropriate to handle the data set?
2. If there aren't any appropriate methods of transforming the data, how can it be modeled? (using Fit Model, etc)
3. After the data is transformed, how do I interpret prediction profiler results if they've been trasformed?
