Questions about the childaggression multiple regression example (DSUR 7 Smart Alex task 4)

Home Forums Default Forum Questions about the childaggression multiple regression example (DSUR 7 Smart Alex task 4)

Viewing 1 post (of 1 total)
  • Author
    Posts
  • #1500
    Randy Reed
    Member

    Hi All,

    Let me say I have a programming background but am noob at statistics and I’m working through Dr. Fields great book. In following task 4 (end ch. 7), I did the following things: a multiple regression analysis using the parenting and sibling rivalry variables child_lm<-lm(Aggression~Sibling_Aggression+Parenting_Style, childAgg)

    And then one including all variables: child_lm.all<-lm(Aggression~Television+Computer_Games+Sibling_Aggression+Diet+Parenting_Style, childAgg)

    Which showed that all variables are significant except television. Pretty much what Dr. Fields notes say.
    The anova analysis then shows that the second analysis is superior.

    However, this leads to my question: why then wouldn’t you create another model, this time excluding
    Television?

    When I did this, and ran anova on it against the baseline model. F is 10.303 with p <.001,
    as opposed to the model using all variables which when compared to the baseline F =7.0339 with p<.001
    Doesn’t that means the model using the variables (-tv) is superior to all variables?

    My second question is related to this. When I ran an anova on all three models I get the following output
    anova(child_lm, child_lm.top, child_lm.all)

    Analysis of Variance Table
    Model 1: Aggression ~ Sibling_Aggression + Parenting_Style
    Model 2: Aggression ~ Computer_Games + Sibling_Aggression + Diet + Parenting_Style
    Model 3: Aggression ~ Television + Computer_Games + Sibling_Aggression +
    Diet + Parenting_Style Res.Df RSS Df Sum of Sq F Pr(>F)
    1 663 64.230
    2 661 62.288 2 1.94180 10.2955 3.954e-05 ***
    3 660 62.240 1 0.04817 0.5108 0.4751

    Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

    which seems to show that the model without television is significant and not the all variables model.
    However when I reverse the positions of model 2 and model 3, It shows the all variables model
    significant and the model without television not. Why the difference?

    I suspect I do not completely understand what I am looking at and that’s why I’m confused in both cases.
    can someone help me?
    Thanks

Viewing 1 post (of 1 total)
  • The forum ‘Default Forum’ is closed to new topics and replies.