Skip to main content

Table 1 Maximum likelihood estimates for the Browne et al [1] two-factor, and the progressive impact, models

From: Structural equation model testing and the quality of natural killer cell activity measurements

  Browne et al.[1] Model ++   Progress ive Impact Model
  NK Activity Factor IFN Activity Factor Indicator R2+++   NK Activity Factor IFN Activity Factor NK Progressive Factor IFN Progressive Factor Indicator R2 Proportion of indicator variance explained by NK Activity Factor Proportion of indicator variance explained by IFN Activity Factor
NK100 .842 .003 .709   .705** -- -80.1** -- .958 .50 --
NK50 .936 -.005 .876   .851** -- -50.0+ -- .918 .72 --
NK25 .943 .015 .892   .920** -- -25.0+ -- .874 .85 --
NK12 .964 -.013 .927   .922** -- -12.5+ -- .995 .98 --
IFN50 .030 .942 .893   -- .897** -- -50.0+ .972 -- .80
IFN25 -.019 .996 .988   -- .977** -- -25.0+ .996 -- .95
IFN12 .005 .995 .990   -- .988** -- -12.5+ .988 -- .98
IFN6 -.018 .991 .977   -- .944** -- -6.25+ .990 -- .99
Factor Variance 1.0+ 1.0+    1.0+ 1.0+ .000069§** .000069§**    
  1. + a fixed coefficient
  2. * beyond 2 standard errors
  3. ** beyond 3 standard errors
  4. § constrained to be equal
  5. ++ Identifying Browne et al's [1] exploratory two-factor model requires excluding one indicator from loading on each factor. Repeated emails to both Browne and MacCallum were unable to elicit a statement of precisely which two loadings had been set to zero. There are 16 different ways of excluding one IFN indicator from loading on the NK activity factor and simultaneously excluding one NK indicator from the INF activity factor (see dashed loadings in Figure 1). The loadings in column 1 and 2 are the average of the estimated loadings calculated across these 16 exclusion possibilities. Each of the 16 models provided a χ2 = 103.8, df = 13, p = 0.000, with the slight difference from the reported fit being easily attributed to the three figure accuracy of the correlations published in Browne et al.[1].
  6. +++ These values are for the version of the Browne et al. [1] model that excludes the effects leading from the NK latent to IFN6 and from the IFN latent to NK100 for identification of the model.