Abstract
The maximum likelihood (ML) estimate and the restricted or residual maximum likelihood (REML) estimate are considered for a common correlation coefficient among several bivariate normal distributions with different variances when some observations on either of the variables are missing. The use of incomplete data in ML and REML estimation reduces mean squared errors of the correlation estimates. Reduction is large when the absolute value of a common correlation is large or numbers of paired observations are small. An example and some simulation results are given to illustrate the characteristics of the estimates.
| Original language | English |
|---|---|
| Pages (from-to) | 1136-1146 |
| Number of pages | 11 |
| Journal | Biometrics |
| Volume | 54 |
| Issue number | 3 |
| DOIs | |
| Publication status | Published - 1998 Sep |
| Externally published | Yes |
Fingerprint
Keywords
- Asymptotic variance
- Fisher information matrix
- Maximum likelihood estimate
- Re-stricted maximum likelihood estimate
- Variance-stabilizing transformation
ASJC Scopus subject areas
- Agricultural and Biological Sciences(all)
- Public Health, Environmental and Occupational Health
- Agricultural and Biological Sciences (miscellaneous)
- Applied Mathematics
- Statistics and Probability
Cite this
Estimation for a common correlation coefficient in bivariate normal distributions with missing observations. / Minami, Mihoko; Shimizu, Kunio.
In: Biometrics, Vol. 54, No. 3, 09.1998, p. 1136-1146.Research output: Contribution to journal › Article
}
TY - JOUR
T1 - Estimation for a common correlation coefficient in bivariate normal distributions with missing observations
AU - Minami, Mihoko
AU - Shimizu, Kunio
PY - 1998/9
Y1 - 1998/9
N2 - The maximum likelihood (ML) estimate and the restricted or residual maximum likelihood (REML) estimate are considered for a common correlation coefficient among several bivariate normal distributions with different variances when some observations on either of the variables are missing. The use of incomplete data in ML and REML estimation reduces mean squared errors of the correlation estimates. Reduction is large when the absolute value of a common correlation is large or numbers of paired observations are small. An example and some simulation results are given to illustrate the characteristics of the estimates.
AB - The maximum likelihood (ML) estimate and the restricted or residual maximum likelihood (REML) estimate are considered for a common correlation coefficient among several bivariate normal distributions with different variances when some observations on either of the variables are missing. The use of incomplete data in ML and REML estimation reduces mean squared errors of the correlation estimates. Reduction is large when the absolute value of a common correlation is large or numbers of paired observations are small. An example and some simulation results are given to illustrate the characteristics of the estimates.
KW - Asymptotic variance
KW - Fisher information matrix
KW - Maximum likelihood estimate
KW - Re-stricted maximum likelihood estimate
KW - Variance-stabilizing transformation
UR - http://www.scopus.com/inward/record.url?scp=0031712556&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0031712556&partnerID=8YFLogxK
U2 - 10.2307/2533863
DO - 10.2307/2533863
M3 - Article
AN - SCOPUS:0031712556
VL - 54
SP - 1136
EP - 1146
JO - Biometrics
JF - Biometrics
SN - 0006-341X
IS - 3
ER -