VSE knjižnice (vzajemna bibliografsko-kataložna baza podatkov COBIB.SI)
  • Grafični prikazi učinkov v regresijskih modelih = Graphical displays of effects in regression models
    Ružić Gorenjec, Nina, 1986-
    Regression models with multiple independent variables are frequently used in statistical data analysis. Independent variables are often included with non-linear effects, and interactions between ... independent variables are added to the model. Because it is usually not possible to present these effects numerically in an intuitive way, graphical presentations are necessary to properly illustrate these effects in reports and scientific papers. For a chosen regression model, we estimate the predicted value of the outcome or its transformed version and plot it together with the corresponding confidence intervals. Independent variables that are included in the effect that we are presenting vary, while all other variables in the model are fixed to certain values. In the paper, we present how the values for which the other variables are adjusted influence the width of the confidence intervals for predicted values of the outcome, implying that graphical inspection of statistical significance of non-linearity of an effect based on confidence intervals is not sensible. Furthermore, we show that confidence intervals in graphical presentations of interactions between independent variables can be misleading, since the (lack of) intersection between them can give wrong intuition about the (lack of) statistical significance of the interaction. In the second part of the paper, we concisely present the rms library from R software, which is very useful for graphical presentations of effects in various regression models. Source code for data simulation and all graphical presentations from the paper is freely available online.
    Vir: Informatica medica slovenica. - ISSN 1318-2129 (Letn. 25, št. 1/2, 2020, str. 19-32)
    Vrsta gradiva - članek, sestavni del ; neleposlovje za odrasle
    Leto - 2020
    Jezik - slovenski
    COBISS.SI-ID - 51683843

    Povezava(-e):

    https://ims.mf.uni-lj.si/ims_archive/25/25-03.pdf
    Digitalna knjižnica Slovenije - dLib.si

vir: Informatica medica slovenica. - ISSN 1318-2129 (Letn. 25, št. 1/2, 2020, str. 19-32)
loading ...
loading ...
loading ...

Vnos na polico