Login  Register

Re: Standardized vs. Adjusted Standardized Residuals for Statistically Significant Chi-Square

Posted by Mark Miller on May 28, 2014; 6:38pm
URL: http://spssx-discussion.165.s1.nabble.com/Standardized-vs-Adjusted-Standardized-Residuals-for-Statistically-Significant-Chi-Square-tp5717593p5726266.html

Marc,

They are NOT the same.
 
 You can transform a residual dividing it by the square root of the 
expected value. This produces the standardized residual, also 
called Pearson residual. In turn, a Pearson residual can be divided by 
the standard deviation of all residuals, thus obtaining the adjusted 
residual. 

The great usefulness of adjusted residuals is that they are standardized 
values, so it is legitimate to compare residuals from different cells. 
Furthermore, adjusted residuals follow a standard normal frequency 
distribution (with mean zero and standard deviation one), so we can use 
a computer program or a probabilities table to come up with the 
probability that a certain residual’s value is not due to chance. In a 
normal distribution, 95% of the values are roughly within the mean plus 
or minus two standard deviations. So, if the adjusted residual’s value 
is greater than two or lesser than minus two, the probability that this 
value is due to chance will be less than 5% and we’ll be able to say 
that the residual is significant 

Adjusted residuals allow us to assess the significance in each cell but, 
if we want to know if there’s a global association between variables we 
have to sum up all adjusted residuals. This is because the sum of 
adjusted residuals also follow a frequency distribution, but this time 
it’s a chi-square frequency distribution with (rows-1) x (columns-1) 
degrees of freedom. 

As far as I know, ADJUSTED residuals were introduced and recommended 
by Shelby J. Haberman in or around 1972 (and thereafter), 
but they have been recommended by many others over the years.
Look at contingency table literature for examples.
The list of possible references is exceedingly long.

Haberman is still an active contributor to this literature (now at ETS).
Cites to Haberman's early work might include

1970: The general log-linear model. Shelby J. Haberman.
    Ph.D. Dissertation, Univerity of chicago.
1972: Algorithm AS 51: Log-Linear Fit for Contingency Tables, S. J. Haberman
Journal of the Royal Statistical Society. Series C (Applied Statistics)
Vol. 21, No. 2 (1972), pp. 218-225

1974: The Analysis of Frequency Data. by Shelby J. Haberman;
Chicago: University of Chicago Press.

... Mark Miller


On Tue, May 27, 2014 at 7:54 PM, marc <[hidden email]> wrote:
Many thanks for this Mark, very helpful.  The link that you posted is the one
that i had previously found, supporting my sense that there's not a lot out
there on this!

One simple query then: 'adjusted standardised residuals' and 'adjusted
residuals' are the same thing?  The 'standardised' is redundant/assumed and
so sometimes included, other times not?



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Standardized-vs-Adjusted-Standardized-Residuals-for-Statistically-Significant-Chi-Square-tp5717593p5726233.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD