Comments - Regression Analysis using R explained - AnalyticBridge2020-08-03T10:15:20Zhttps://www.analyticbridge.datasciencecentral.com/profiles/comment/feed?attachedTo=2004291%3ABlogPost%3A317210&xn_auth=noAs you mentioned regression h…tag:www.analyticbridge.datasciencecentral.com,2015-01-02:2004291:Comment:3175622015-01-02T02:54:55.133ZKhurramhttps://www.analyticbridge.datasciencecentral.com/profile/Khurram
<p>As you mentioned regression helps answers to find association, relationship between variables.As per my understanding association relates to correlation where variables completly not dependent to another variable. This is true with regression which let you know the relation between variable.</p>
<p>As you mentioned regression helps answers to find association, relationship between variables.As per my understanding association relates to correlation where variables completly not dependent to another variable. This is true with regression which let you know the relation between variable.</p> Note that the assumptions on…tag:www.analyticbridge.datasciencecentral.com,2014-12-30:2004291:Comment:3173752014-12-30T11:53:24.049ZProf. Dr. Diego Kuonenhttps://www.analyticbridge.datasciencecentral.com/profile/DiegoKuonen
<p><span>Note that the assumptions on the errors of the multiple linear regression model are not satisfied!</span><br></br><br></br><span>For example, huge residuals clearly demand a robust fit (e.g. using MM-estimation as in "lmRob" of R package "robust", or "lmrob" of package "robustbase"), and ACF of resid. may show autocorrelations with such dependent (time-ordered) data (e.g. using "acf(resid(model2))" in R.</span><br></br><br></br><span>It makes sense to base inferences or conclusions only on valid…</span></p>
<p><span>Note that the assumptions on the errors of the multiple linear regression model are not satisfied!</span><br/><br/><span>For example, huge residuals clearly demand a robust fit (e.g. using MM-estimation as in "lmRob" of R package "robust", or "lmrob" of package "robustbase"), and ACF of resid. may show autocorrelations with such dependent (time-ordered) data (e.g. using "acf(resid(model2))" in R.</span><br/><br/><span>It makes sense to base inferences or conclusions only on valid models. In other words, any conclusion is only as sound as the model on which it is based.</span></p> Thanks Justice Moses for the…tag:www.analyticbridge.datasciencecentral.com,2014-12-29:2004291:Comment:3174552014-12-29T10:07:22.499Zsuresh kumar Gorakalahttps://www.analyticbridge.datasciencecentral.com/profile/sureshkumarGorakala
<p>Thanks Justice Moses for the explanation. Will takecare of such things in future</p>
<p>Thanks Justice Moses for the explanation. Will takecare of such things in future</p> Hi Suresh,
Many thanks for th…tag:www.analyticbridge.datasciencecentral.com,2014-12-27:2004291:Comment:3174412014-12-27T20:39:47.043ZJUSTICE MOSES K. AHETOhttps://www.analyticbridge.datasciencecentral.com/profile/JUSTICEMOSESKAHETO
<p>Hi Suresh,</p>
<p>Many thanks for throwing more light on some basics of regression models/analysis, well done.</p>
<p>The general regression model formula you presented at the top of the graph is correct.</p>
<p>However, the estimated regression model below the graph is quiet not right. Once you introduced the cap at the top of y (y with cap on top), you now have a fitted regression model whose expected error must be zero, ie E(e)=0. This means that you should have <span>ˆy = ˆ β0 + ˆ β1x…</span></p>
<p>Hi Suresh,</p>
<p>Many thanks for throwing more light on some basics of regression models/analysis, well done.</p>
<p>The general regression model formula you presented at the top of the graph is correct.</p>
<p>However, the estimated regression model below the graph is quiet not right. Once you introduced the cap at the top of y (y with cap on top), you now have a fitted regression model whose expected error must be zero, ie E(e)=0. This means that you should have <span>ˆy = ˆ β0 + ˆ β1x instead of <span>ˆy = ˆ β0 + ˆ β1x + ˆe </span></span>unless in multilevel model in which apart from allowing the observations at higher level to be correlated within the group, you are also allowing for complex level 1 residuals. </p>
<p>Hope this helps.</p>
<p>Once again, well done. </p>