Skip to main content

Advertisement

Table 5 Performances of methods using instrument pr

From: Instrumental variable analysis in the context of dichotomous outcome and exposure with a numerical experiment in pharmacoepidemiology

   Instrument strength
   Weak Mod Strong
Level Method rB sd rMSE pval rB sd rMSE pval rB sd rMSE pval
High Tr 1.62 0.23 0.24 0.06 0.52 0.26 0.24 0.04 1.93 0.40 0.32 0.06
  Conv 46.14 0.23 1.40 1.00 42.40 0.26 1.29 1.00 39.62 0.39 1.23 1.00
  2SRI 14.98 0.66 0.85 0.13 9.41 0.47 0.53 0.09 8.49 0.44 0.53 0.10
  GMM 72.01 315.65 5.06 0.27 75.50 109.91 4.70 0.32 71.32 60.00 4.25 0.40
Med Tr 0.58 0.25 0.23 0.05 1.25 0.31 0.30 0.07 1.61 0.48 0.32 0.04
  Conv 25.40 0.25 0.80 0.96 23.98 0.31 0.78 0.85 21.37 0.48 0.72 0.63
  2SRI 9.93 0.66 0.70 0.07 6.90 0.54 0.61 0.08 5.45 0.54 0.54 0.09
  GMM 36.76 47.90 43.89 0.40 32.85 32.04 2.17 0.51 44.82 41.87 3.36 0.35
Low Tr 0.66 0.26 0.26 0.07 1.51 0.33 0.28 0.04 2.20 0.55 0.35 0.05
  Conv 7.86 0.26 0.35 0.15 7.96 0.33 0.37 0.11 7.72 0.55 0.41 0.10
  2SRI 6.43 0.71 0.74 0.07 5.76 0.64 0.62 0.06 5.91 0.66 0.67 0.10
  GMM 22.04 31.69 2.42 0.49 26.15 31.68 3.49 0.64 43.12 25.42 4.14 0.46
  1. Legend: Tr = True model, Conv = Conventional model, 2SRI = Two-Stage Residual Inclusion, GMM = Generalized Method of Moment. Low, Med (Medium), High denote the level of confounding whereas Weak, Mod (Moderate), Strong stand for instrument strength. For the criteria, rB= relative bias (%),sd= standard deviation, rMSE= root Mean Squares error and pval= non-coverage probabilities