Go Back to
Jeremy Miles's Home page
Books by Jeremy Miles
Regression Analysis Home page
About "Applying Regression and Correlation"

Errors and Omissions

Jump straight to:
Errors
Omissions

 

Errors (that have been found so far):  

On page 6, right below the first SD calculation,  did not include the answer: SD = 1.73 (above the next sentence: "Similarly, ...).
Page 17, The text (line 2, after the first upper and lower CI equations) reports that "would yield values between 3.22 and 8.97".  This does not match the calculations - it shoud say ""would yield values between 2.51 and 8.97".
Page 29, Page 35: (Only of interest if you want to download the data files, which you can do here .)  The data in Table 2.1is labelled as dataset 2.1.  If you go to the downloads page you will find it is dataset 1.1.  The data in Table 2.7 is labelled as dataset 2.2.  If you go to the downloads page you will find it is dataset 2.1. 
Page 129, Table 5.4, :  Column 1 should be DV and Column 2, IVs.  (Not, as is written, the other way around.
Page 130, line 3, "independent variable" should be "dependent variable".
Page 151, paragraph 4.  "(R^2 = 0.097, F(1, 48)=5.1, p < 0.028), should be "p = 0.028".
Page 161, line 10, missing minus sign, should say "odds=exp(-0.84) = 0.43".
Page 184,in table 7.16, the F associate with model 2 should be 8.073, not 9.073.

Page 185, equation below table 7.17: on the first line, substitute zbooks for zgrade.  Should say:
grade = (5.95 x zbooks) + (5.70 x zgrade) + 4.50 x zbooks x zgrade)

Page 223, paragraph 1.  Not really a mistake, but it isn't very clear, is it?  It says:

The first thing to do is to add the mean of the y variable to the constant.  When we do this the line will no longer hit the y-axis at zero, but will hit it at the mean value of Y.   But this value now gives the value of the intercept if the y-axis crossed the x-axis at the mean value of x, not at zero. So we need to correct for the mean of x, but unless the slope is equal to 1, but we need to find out how much the line changes between the mean of x, and the point at which x = 0.   To do this we use the slope, because the change must be equal to the mean of x multiplied by the slope.  

Substitute something like:

The constant (intercept) is currently zero.  .  If we add the mean of y to the constant (which, remember is currently zero), the constant will become the mean of y.  This means that the slope now would intercept the y-axis, if the y-axis were placed at the mean of x .   (We are using the centred version of the x variable, so the mean is zero). We want the intercept to be corrected, so that it hits the y-axis when the y-axis is placed at zero on the x-axis.  We need to move (mean(x)) units down the slope.  But we need to know how far this will lower the intercept.
If we moved 1 along x, we drop b units on y.  So if we move mean(x) units on the x-axis, we move b * mean(x). 



Page 30 says: "...the standardised regression coefficient ... between the two variables is 0.482, This correlation of 0.482 is lower than the standardised regression coefficient between books and grade (which was 0.492), and so we might say that attendance is a better predictor of final grade in a course than is the number of books read..."

Of course this is wrong.  The opposite is true - it should say "Number of books read is a better predictor of final grade than attendance."
On page 37 the first sentence in the last paragraph should read "We also find that the numerator df is equal to k2-k1=3-2=1 (instead of 2-1=1) and the denominator is equal to N-k2-k1=40-3-2=35 (instead of 40-3-1=36).
On page 222, the formula for calculating a slope should be b=rxy*SD(y)/SD(x) (instead of SD(y) in the numerator and SD(x) in the denominator).


Thanks to:
Miles Cox, University of Wales, Bangor, Wales.
Ellen Mastenbroek, Leiden University, The Netherlands.
Ine Deprouw, Artesia Bank, Brussels, Belgium.
Stephen Gay, Children's Hospital Medical Center of Cincinnati, Ohio, USA.
Marie Evertsson, Swedish Institure for Social Research, Stockholm University.
Eric Melse, Maastricht University, The Netherlands.
Charles Meliska,  University of California, San Diego, USA.

Found anything else wrong?  Please see contacts below.

 

Omissions
Chapter 2: We should really have mentioned something on suppressor variables here, but didn't.  These are a complex interplay between IVs, in which including some IVs in a regression equation can enhance, or reverse, the effect of other IVs.  I will get around to writing something on this, and will put it here when I do.  If you want me to hurry up, send me an email . 
In the second edition, we will add more on different types of correlation, e.g. Kendall's, Spearmans.

Page 190.  We mention one way of calculating the amount, or degree, of mediation.  We could have mentioned that this is also equal to the effect of X on M, multiplied by the effect of M on Y.  In the example:

Effect of X (enjoy) on M (buy) = 0.974 (from step 2).
Effect of M (buy) on Y (read) = 0.206 (from step 3).

Mediation effect = 0.974 x 0.206 = 0.200.

This is the same as the result of subtracting the effect of step 4 from step 1.

We could also have told you that it is possible to calculate a standard error (and hence a confidence interval, t-value and significance) for a mediated effect.  The calculations go beyond what we covered, but can be found on David Kenny's web page .

 
Anything you would like to be in here?  Send me an email? 

 


FREE HIT COUNTER

Contact

Got a comment on this web site?  Contact webmaster@jeremymiles.co.uk
Got a comment on the book?  Contact jeremy@jeremymiles.co.uk