Dealing with missing data in a multi-question depression scale: a comparison of imputation methods

Table 4 Correlation with increased missing items in the P = 0.20 missing data scenario.

Method of Imputation	Random Selection		Preceding Question		Question Mean		Individual Mean		Single Regression		Multiple Imputation
# of Missing Items (P = 0.20)	Spearman	Kappa	Spearman	Kappa	Spearman	Kappa	Spearman	Kappa	Spearman	Kappa	Spearman	Kappa
1 (N = 83)	0.983	0.971	0.983	0.910	0.993	1.000	0.992	0.971	0.990	0.970	0.993	1.000
2 (N = 220)	0.941	0.605	0.961	0.741	0.987	0.804	0.987	0.794	0.984	0.783	0.988	0.841
3 (N = 341)	0.877	0.563	0.944	0.791	0.981	0.771	0.980	0.833	0.974	0.832	0.982	0.848
4 (N = 314)	0.816	0.454	0.916	0.739	0.976	0.743	0.973	0.823	0.967	0.764	0.974	0.854
5 (N = 296)	0.783	0.438	0.930	0.661	0.978	0.641	0.977	0.780	0.969	0.784	0.978	0.853
6 (N = 171)	0.618	0.182	0.890	0.569	0.963	0.651	0.953	0.745	0.934	0.710	0.957	0.805
7 (N = 92)	0.508	0.114	0.791	0.454	0.948	0.346	0.930	0.630	0.918	0.600	0.948	0.649
8 (N = 34)	0.563	0.191	0.848	0.698	0.970	0.401	0.958	0.827	0.943	0.884	0.978	0.831

ISSN: 1471-2288