Thanks a lot Phil for this video. It is very helpful, instructional and detailed. It's all we need as a starter in multiple regression and for understanding the important basics.
By far, this is the best video I have ever gone through with regard to linear regression. Awesome presentation and right to the point. In deed helpful for people at any stage of their learning, of SPSS statistical methods. !
I logged in just to tell you this: By far the simplest and clearest explanation I've had on this. Why oh why do lecturers attempt to 'pomp' up these matters. Doesn't science teach us about parsimony!? Anyway...much appreciated. You might think that it's a simple youtube video, commented on by some random viewer, but (not hyperbole) you changed my life today. Thank you. Thank you. Thank you.
Glad it was helpful. I think the question you ask is answered in this other video of mine called "SPSS for newbies: Why having a high R-squared in regression could be a bad thing "
hey Phil, to get rid of the yellow box that appears every time you hoover your mouse over the tables, you can copy the table to a word document and explain the tables there instead of SPSS interface. :)
It was good. I would like to know if the R found in the example is enough to consider the model as strong, and what are de intervale to decide about that.
Hi. If you are a newbie/taking an intro course in modelling, then you are likely to be taught that since the coeff is not significant, you may try removing it an refitting. In reality, it's not so straight forward as the choice depends on numerous things like: reason for building a model, results of other (diagnostic) tests. But you can ignore my last comment if you are a newbie!
Hi Phil, Thank you for this video! It is very helpful. What does the (Constant) represent? I am currently running a linear regression model in which other independent variables (including the total model) is significant, however the (constant) is not.
+Leo Christian Navarra Leo - the standardized coeff allows you to assess the relative importance of each IV to the others. To rank the importance of the Xs, drop any minus symbol in front of the coeff then compare the numbers.
Thanks for the video and guidance! But I have a question: I did a multiple regression analysis and the results showed that there were significant correlation between the DV and IVs, however in the coefficient table certain IVs had p-value higher than 0.05. How do I interpret this?
If I understand correctly, you looked at correlations between DV and IVs and found that some were significant, but then when you put them in the regression the coeff was insignificant. This is common place when running multiple regression. From a newbie perspective this could be because of multicollinearity problem (I have a video explaining this). Other reasons that could explain what you see in your analysis are mediation, and confounding variables. These last 2 need another video, which I have in the pipeline.
Hi Phil, thank you so much for this video, it really helps a lot. I just have one question, at the end of the video you say that the F and T statistics are valid if certain conditions hold for the model. Can you please list these conditions or point me towards material teaching about these conditions. Or maybe you have another video where you talk about this conditions. Thanks
hey phil, first of all, big big thanks for ur great videos of spss.. what are the explanations between simple linear regression and multiple linear regression? could u pls give some examples? do reply plzzz.. know my questions are idiotic, but i am really now out of help now.. u'll never imagine how excited i m now running into your videos. really life saver..(thumb up)
Hello, This video is helpful, one more question: What is the meaning of the p value of the constant (0.029) in the last table? What if its > 0.05? it means the model is not significant or what? because I have two IVs, one is significant (.001) and another is not significant (.689) and the p value of the top column is .332, what is it stand for? Thanks
That's for the intercept. p-value for intercept >0.05 suggests it's not significant, but do not delete it. In applications, the importance on the slope not the intercept.
Hi! Great video! However, I have a question regarding significance. I applied MLRM using 5 independent variables with adjusted r-squared 0.820. Even though the r-squared, one can say, that is acceptable, the significance (sig.) in all independent variables is way higher than 0.05 (e.g. 0.018, 0.508, 0.682). How do I decide whether to keep or reject the model? Thank you in advance!
Rash - that you have so many Xs insignifcant yet high adjusted R-square points to a problem of high collinearity (mutlicollinearity). Look it up in your lecture notes? Good luck.
Thank you very much for your prompt reply. I don’t have any notes, since I am self-taught in statistical analysis. Briefly, I have a value (dependent variable) that is related to other 5 parameters (independent variables). The correlation and linear regression each of the independent variables with the dependent variable is good (r>0.8, p
What newbies are taught is to build a model that satisfy the NICE-L assumption: Normally distributed error term, Independent errors (ie random sampling, or test if it's ordered data), Constant variance of error, Expectation of error is zero, Linear in parameters. There are a whole load of stats and plots to examine these things, so it's not a case of basing your decision on 2 stats - R^2 and individual sig - you take the stats together as a whole and interpret them together. Interpretation is rather subjective - there is no one correct answer - we are not doing pure math. Best to get a book.....I am making a series on Regression for newbies - check out my playlist.
Thank you so much for this! If my coefficient is very small (-0.004), how can I interpret that or how should I move the decimal? I hope this makes sense...
Supposing you are talking about a slope parameter Think about the units of measurement your x that has the small param. A rescaling of the units of X would change the param. eg if X were measured in cents then changing units to 1000 dollars would change it to -4 from -0.004.
Phil Chan, i don't know if you are still using this youtube account but if anyone can help me:I have found a significant value in the ANOVA table (0.024) but not any significant values in the coefficients table (0.054 and 0.132) what does this say?
Sir thank you! I have a question...if the whole model is rejected (ANOVA) considering the P-value then subsequent coefficients are also useless to be explained. My understanding we will explain and discuss about coefficients only when model is found to be fit with explanatory power... Kindly discuss it
The null hypothesis in F-test in the Anova table is that the model has no explanatory power for Y. If you reject this, and generally one does reject it, then you may have a look at the coefficients. Note, one should test the assumptions of regression (NICE-L assumption) before looking at the F and t tests.
By the way you phrase your question, I wonder whether you understand the F-test. When you say the overall model is rejected, I take this to mean you reject to null of this F-test, and this means there is evidence the model has explanatory power. BUT perhaps you mean to say that suppose we DO NOT reject the F-test ie no evidence the model has explanatory power, then do the coeff have meaning. Supposing all the assumptions underpinning the model are satisfied then if there is no evidence the model has explanatory power those coeff don't tell us anything. Good luck.
+Phil Chan what does it mean if the constant in the coefficient table is non significant, but everything else in that table is significant? How do I interpret that?
+Klaudia C Klaudia - simple answer is to keep it in the model and ignore it. After all, the purpose of a regression model is for prediction (of the dependent variable) or interpretation of the coefficients on the predictors (Xs). For both, we don't have to be concerned with the intercept (the "constant" in SPSS).
I see. That's 2 different things - the overall significance of a model (F-test in anova table), and partial F-test for assessing significance of new Xs added in (ie the change in the R2). Certainly it is plausible that the partial F-test does not reject the null, but the 1st test of overall significance rejects the null. You can look them up somewhere that explains what they are, but they should not be confused: F-test for overall sig, and partial F-test
Hello. Could you please help me. I made 4 different variables regarding recycling (paper, glass, plastic, metals). The values are 1=yes, 2=0. I created multiple response set afterwards and got the result but, how can i check how many people recycle all four materials? Thank you
what if my p values are more than .05 in the coefficients table? Doesn't that mean that I cannot reject the null Hypothesis and that my variables do not show any significance and cannot predict my dependent variable?
+Robin h Robin - if you are after a model that aims to explain the CAUSAL relationship between Y and IVs then as you say a p-value > 0.05 (used as standard) on a coeff means the corresponding IV does not explain the Y. On the other hand if you are out to build a model to PREDICT Y then these p-values don't matter. Unless you are a major in the computer science/electronic engineering, it's likely the emphasis of your course is to build a model for explaining the causal relationship.
+Phil Chan (statisticsmentorcom) Thank you!! I also watched your video on the chi square test. What exactly is the difference between the two tests? Is chi square just looking to see if the variables are associated and the linear regression is trying to predict if the variables influence each other? I have to use both tests for a paper and am confused on how they differ.
another question please if i have F=1.929 and the sig is 0.044 what do both tell us? sorry for disturbing but i am trying to understand the output. thank you😊