Abstract
The purpose of this paper is to solve the problem of multicollinearity that affects the estimation of logistic regression model by introducing first-order approximated jackknifed ridge logistic estimator which is more efficient than the first-order approximated maximum likelihood estimator and has smaller variance than the first-order approximated jackknife ridge logistic estimator. Comparisons of the first-order approximated jackknifed ridge logistic estimator to the first-order approximated maximum likelihood, first-order approximated ridge, first-order approximated r-k class and principal components logistic regression estimators according to the bias, covariance and mean square error criteria are done. Three different estimators for the ridge parameter are also proposed. A real data set is used to see the performance of the first-order approximated jackknifed ridge logistic estimator over the first-order approximated maximum likelihood, first-order approximated ridge logistic, first-order approximated r-k class and first-order approximated principal components logistic regression estimators. Finally, two simulation studies are conducted in order to show the performance of the first-order approximated jackknife ridge logistic estimator.
Similar content being viewed by others
Notes
\(V(\beta _{0})\) will be denoted as \( V^{(0)}=diag({\hat{\pi }}_{i}^{(0)}(1-{\hat{\pi }}_{i}^{(0)}))\) in the following sections where \({\hat{\pi }}_{i}^{(0)}\) is evaluated at \(\beta _{0}\).
There is no global mechanism for good starting values. Therefore, starting value \({\widehat{\beta }}^{(0)}\) should be known which can equivalently be taken as the real parameter value \(\beta _{0}\). \({\widehat{\beta }} ^{(0)}=\beta _{0}\) is usually the ordinary least squares estimate in practice (see for example Schaefer 1986).
LeCessie and VanHouwelingen (1992) defined a one-step approximation for \( {\hat{\beta }}_{-i}(k)\) as \({\widehat{\beta }}_{-i}^{(1)}(k)={\widehat{\beta }} ^{(1)}(k)-\frac{(X^{\prime }{\widehat{V}}(k)X+kI)^{-1}X_{i}^{\prime }v_{ii}(y_{i}-{\widehat{\pi }}_{i})}{1-h_{ii}(k)}\) where \({\widehat{V}}(k)\), \( {\widehat{\pi }}_{i}\) and \(h_{ii}(k)\) are evaluated at \({\widehat{\beta }}^{(1)}(k)\). However, their approximation does not yield the jackknife estimator.
From Eq. (1), \({\widehat{\alpha }}^{(1)}\) equals \({\widehat{\alpha }}^{(1)}= {\widehat{\alpha }}^{(0)}+(Z^{\prime }V^{(0)}Z)^{-1}Z^{\prime }(y_{i}-{\hat{\pi }}_{i}^{(0)})\) where \({\widehat{\alpha }}^{(0)}\) is the starting value of \( \alpha \). To start the iteration \({\widehat{\alpha }}^{(0)}\) should be known. Therefore, a real parameter, say \(\alpha ^{0}=T^{\prime }\beta _{0}\), can be used.
All comments, results and views in this study belong to the author(s) and these do not bind TURKSTAT.
OECD equivalence scale is used to reflect differences in a household size and composition. This scale gives a weight to all members of the household (and then adds these up to arrive at the equivalised household size): 1 to the first adult; 0.5 to the second and each subsequent person aged 14 and over; 0.3 to each child aged under 14 (see EUROSTAT 2014).
The sub-indices e and j respectively denote the exact and jackknife CIs.
References
Aguilera AM, Escabias M, Valderrama MJ (2006) Using principal components for estimating logistic regression with high-dimensional multicollinear data. Comput Stat Data Anal 50:1905–1924
Bull SB, Hauck WW, Greenwood CMT (1994) Two-step jackknife bias reduction for logistic regression mles. Commun Stat Simul Comput 23(1):59–88
Bull SB, Shelley B, Greenwood CMT, Hauck WW (1997) Jackknife bias reduction for polychotomous logistic regression. Stat Med 16(5):545–560
EUROSTAT (2014) Living conditions in Europe. Publication Office of the European Union, Belgium
Farebrother RW (1976) Further results on the mean square error of ridge regression. J R Stat Soc B 38:248–250
Gruber MHJ (1991) The efficiency of jack-knifed and usual ridge type estimators: a comparison. Stat Probab Lett 11(1):49–51
Hinkley DV (1977) Jackknifing in unbalanced situations. Technometrics 19(285):292
Kibria BMG (2003) Performance of some new ridge regression estimators. Commun Stat Simul Comput 32(2):419–435
LeCessie S, VanHouwelingen JC (1992) Ridge estimators in logistic regression. Appl Stat 41(1):191–201
McDonald GC, Galarneau DI (1975) A monte carlo evaluation of ridge-type estimators. J Am Stat Assoc 70:407–416
Nelder JA, Wedderburn WM (1972) Generalized linear models. J R Stat Soc A 135(3):370–384
Nomura M (1988) On the almost unbiased ridge regression estimator. Commun Stat Simul Comput 17:729–743
Nyquist H (1988) Applications of the jackknife procedure in ridge regression. Comput Stat Data Anal 6(2):177–183
Ogoke UP, Nduka EC, Nja ME (2013) The logistic regression model with a modified weight function in survival analysis. Math Theory Model 3(8):12–17
Ohtani K (1986) On small sample properties of the almost unbiased generalized ridge estimator. Commun Stat Theory Methods 15:1571–1578
Özkale MR (2008) A jackknifed ridge estimator in the linear regression model with heteroscedastic or correlated error. Stat Probab Lett 78:3159–3169
Özkale MR, Arıcan E (2015) First-order r-d class estimator in fbinary logistic regression model. Stat Probab 106:19–29
Özkale MR, Arıcan E (2016) A new biased estimator in logistic regression model. Statistics 50(2):233–253
Özkale MR, Lemeshow S, Sturdivant R (2018) Logistic regression diagnostics in ridge regression. Comput Stat 33:563–593
Pregibon D (1981) Logistic regression diagnostics. Ann Stat 9(4):705–724
Quenouille MH (1956) Notes on bias in estimation. Biometrika 43:353–360
Rao CR (1973) Linear statistical inference and its applications, 2nd edn. Wiley, New York
Schaefer RL (1983) Bias correction in maximum likelihood logistic regression. Stat Med 2:71–78
Schaefer RL (1986) Alternative estimators in logistic regression when the data are collinear. J Stat Comput Simul 25:75–91
Schaefer RL, Roi LD, Wolfe RA (1984) A ridge logistic estimator. Commun Stat Theory Methods 13:99–113
Singh B, Chaubey YP (1987) On some improved ridge estimators. Stat Pap 28:53–67
Singh B, Chaubey YP, Dwivedi TD (1986) An almost unbiased ridge estimator. Sankhya 48:342–346
Tukey JW (1958) Bias and confidence in not quite large samples (abstract). Ann Math Stat 29:614
TurkStat (2011) Income and living conditions survey 2011 micro. Data set (cross-sectional). Ankara
Wu J, Asar Y (2016) On almost unbiased ridge logistic estimator for the logistic regression model. Hacet J Math Stat 45(3):989–998
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Özkale, M.R., Arıcan, E. A first-order approximated jackknifed ridge estimator in binary logistic regression. Comput Stat 34, 683–712 (2019). https://doi.org/10.1007/s00180-018-0851-6
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00180-018-0851-6