Theft and Fraud for Optimization Criteria

1 He(2009)'s Optimization criterion

2 Optimization criterion in Lin(2009)

In Lin(2009) [Biometrika, Volume 96, Issue 1, March 2009, Pages 243–247], C.D.Lin defined an Optimization criterion max, see Figure 2,

3 Optimization criterion in Lin(2008,2010) & Bingham(2009)

The optimization criterion in Lin (2008, 2010) and Bingham (2009) are the same, which is shown in Figure 3,

Figure 3 optimization criterion in Lin (2008, 2010) and Bingham (2009)

4 C.D.Lin theft, lying and fraud

On the issue of optimization criterion, Lin & Tang theft, lying, and fraud.which showed him up completely.

The max in Lin (2009) is the mcc in He (2009), but they dare not copy it completely. The "max" is not a correct definition, but the art of stealing action. "Maximum absolute correlation" has ambiguous semantics, its mathematics and logic are faulty, How to define "absolute correlation", there is no such idiom in mathematics. Correlation should refer to the correlation coefficient, and absolute correlation should refer to the absolute value of the correlation coefficient. "Maximum" must be the largest in a group, and must be the maximum in R defined by them, and it must not be an element on the main diagonal in R (because the value of the element on the main diagonal line is a constant, that Is 1,and the word "main" cannot be omitted). "Max" can only be "the maximum element in absolute value off-main diagonal line of R", which is exactly the mcc in He (2009). They deleted the most important condition from mcc.

They finally proved one thing, they were using max (ie mcc) as the optimization criterion to construct what they called OLHD and NOLH. Lin (2009, 2008, 2010) "published" only this set of OLHD and NOLH, no other.

They falsified and lied in Lin (2008), which further proved their theft.

Lin(2008)(p.24) and Lin(2010)(p.13) used three optimization criterions: A₂=∑_i<jr_ij², then changed "r" in A₂ to ρ, and A₂ into A₂/[m(m-1)/2] , that is ρ²(M) in Bingham(2009); The third was ρ_M.

In Lin (2008), the author said that they did not use max, but ρ²(M), and ρ_M and ρ²(M) were from Bingham (2008). This is a big lie. The whole section 2.4.3 was lying,

Bingham (2008) Suspected Bingham (2009), Its submission date was more than 8 months later than He(2009)'s submission date, and the revised draft was more than 6 months late. The four PhDs came all from the Department of Statistics and Actuarial Science, Simon Fraser University, they are all disciples of Boxin Tang, and close collaborators and share resources. A large part of the third chapter of Lin (2008) was copied from Bingham (2009), and before 2008, Lin cited Bingham's results published in 2009. Can these prove the close cooperative relationship between them? The three authors of Bingham (2009) are the senior supervisor and supervisor of Dr. Lin’s doctoral thesis defense. Therefore, it cannot be ruled out that ρM of Bingham (2009) comes from He (2009), and the general director of this event is Boxin Tang.

The function of the three criterions is the same, measure the correlation. Among them, mcc is the simplest in the form, programming is the easiest, running speed is the fastest and it can directly computing the correlation confidence probability p value; while ρ² is the most complex in the form, programming is the most troublesome and running speed is the slowest, and the p value cannot be calculated. Here is a rough comparison, under the same conditions, if the running time is 1 for mcc, about 2.4 for ρ². If the loop limit is setted 9×10⁶ suggested by Lin, and ρ² is used as the optimization criterion, the running time will be much greater than 10. When runs = 11, the time required for the Lin algorithm is more than 2,000 times the time of the mcc. When runs = 21, the time required for the Lin algorithm is more than 220 times that of the He (2009) algorithm. If He(2009) takes 10 minutes, then Lin algorithm takes 37 hours. Was they use such an algorithm?

In terms of the measurement effect, there was an example in Bingham (2009), in the design, ρ_M(D)= 0.0286, ρ²(D)=0.0286²/(2k-1)= 0.0008 / (2k-1). If takes three decimal places and k=2, then ρ²(D)=0, that is, the design is orthogonal, but mcc=0.0286, P=0.0428, it is an approximate OLHD with a 0.95 confidence probability.

Lin did a comparative study, but she did not report the real experimental data, and only provided results in her favor. This is a typical fraud and deception.