Skip to main content

Table 3 Data A250, B250, and C250. Selected MFP models with complete data (“all”) and after removal of IPs identified from the univariable (IPXu) and multivariable (IPXm) diagnostic analyses. The number of IPs identified univariable and multivariable, respectively, are shown in parentheses. “=” is used if the power selected agreed to the power from all data

From: Effects of influential points and sample size on the selection and replicability of multivariable fractional polynomial models

Type

Variable

Dataset A250 (5, 5)

Dataset B250 (3, 3)

Dataset C250 (4, 6)

True model

All

IPAu

IPAm

All

IPBu

IPBm

All

IPCu

IPCm

Cont.

x1

1

=

=

−1, 3

0, 3

1

Out

=

=

0.5, 1

 

x3

1

=

=

Out

=

=

Out

=

=

1

 

x5

0, 3

−0.5

0

0

=

=

−0.5

=

=

−0.2

 

x6

0

=

=

−0.5

=

=

0

=

=

0

 

x7

Out

=

=

Out

=

=

Out

=

=

Out

 

x10

1

=

3

Out

=

=

1

=

Out

1

Cat.

x9a

Out

=

=

In

=

=

Out

=

=

Out

 

x9b

Out

=

=

Out

=

=

Out

=

=

Out

Bi

x2

Out

=

=

Out

=

=

In

=

=

Out

 

x4a

In

=

=

In

=

=

In

=

=

In

 

x4b

Out

=

=

Out

=

=

Out

=

=

Out

 

x8

In

=

=

Out

=

In

In

=

=

In

  1. Cont, Cat, and Bi denote continuous, categorical, and binary variables respectively