Overview

This week I will work on two objects. One is to validate my code and the other is to find out which algorithm is better in terms of dependent data.

Code validating

If we set up the data with the strength of signals equal to 0, then calculate the power with the same methods as before, the result should be closed to the value we set to control the type one error. The reason is that the distribution of noise and noise plus signals are same, in this condition, the value of type one error will equal to value of power.

  1. reg
##         0      0      0      0      0
## 5  0.0550 0.0750 0.0730 0.0645 0.0580
## 10 0.0645 0.0715 0.0745 0.0590 0.0755
## 15 0.0680 0.0645 0.0655 0.0515 0.0700
## 20 0.0695 0.0440 0.0475 0.0575 0.0485
## 25 0.0650 0.0550 0.0635 0.0625 0.0580
  1. smo (type one error)
##         0      0      0      0      0
## 5  0.0590 0.0485 0.0555 0.0620 0.0595
## 10 0.0565 0.0525 0.0505 0.0595 0.0540
## 15 0.0555 0.0555 0.0570 0.0445 0.0535
## 20 0.0505 0.0510 0.0510 0.0610 0.0560
## 25 0.0540 0.0515 0.0575 0.0600 0.0630
  1. reg (permutation)
##         0      0      0      0      0
## 5  0.0562 0.0598 0.0498 0.0454 0.0566
## 10 0.0528 0.0576 0.0544 0.0550 0.0574
## 15 0.0572 0.0576 0.0556 0.0574 0.0554
## 20 0.0570 0.0588 0.0536 0.0524 0.0492
## 25 0.0616 0.0524 0.0548 0.0560 0.0580
  1. smo (permutation)
##         0      0      0      0      0
## 5  0.0538 0.0562 0.0532 0.0562 0.0514
## 10 0.0572 0.0542 0.0542 0.0510 0.0488
## 15 0.0570 0.0516 0.0528 0.0578 0.0562
## 20 0.0522 0.0548 0.0606 0.0564 0.0564
## 25 0.0532 0.0588 0.0546 0.0518 0.0548

The results are colsed to the value of the type one error we set. Besides, the first method is slightly higher, which is what professor expected. We can say the code is valid.

Later work