LISTSERV at the University of Georgia
Menubar Imagemap
Home Browse Manage Request Manuals Register
Previous messageNext messagePrevious in topicNext in topicPrevious by same authorNext by same authorPrevious page (December 2008)Back to main SPSSX-L pageJoin or leave SPSSX-L (or change settings)ReplyPost a new messageSearchProportional fontNon-proportional font
Date:         Mon, 22 Dec 2008 07:02:29 -0500
Reply-To:     Hannah State-Davey <hmclarke@HOTMAIL.CO.UK>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Hannah State-Davey <hmclarke@HOTMAIL.CO.UK>
Subject:      Comparison of Pearson's and Spearman's Correlation Cofficient


I am constructing a measurement scale and have used a 7-point Likert scale. Due to the lack of agreement over whether Likert scale data can really be considered interval data and the fact that it results in a skewed distribution, I am looking to compare the results of Pearson's and Spearman's correlations on the data. Although I have a large enough sample to reduce the effects of skewness and kurtosis, and the scatterplot reveals a reasonable linear fit to the data (well equally good linear or quadratic fit) I still feel I need to confirm that Pearson's is appropriate to use for this data. The pearson's correlations are primarily larger than spearman's so i guess it is not underestimating the strength of the relationship due to non-linearity as would be indicated if spearman's correlations were larger than pearson's. Is it that the divergence I am seeing between spearman's and pearson's a product of the data (i.e. there will be a lot of ties in the data due to people rating scale items the same)?

I have calculated item-total correlations as the main form of item reduction before the use of factor analysis using both Spearman's and Pearson's. I have ordered the results from both methods highest to lowest and ranked the sets of scores. I then used Spearman's to determine the degree of association between the two ranked sets of scores. I found a high correlation which would indicate that the same set of items would be selected by either of the methods (Nunnally, 1978).

I am also examining the inter-item correlations to determine bloated specifics or items with a very high number of low correlations indicating that they don't measure the same as the other items. Can I do a similar method as above to determine if both correlation methods would result in the same set of items being selected? Is it enough to just do it for item- total correlations? Or am I unduly worrying about the applicability of Pearson's here?

Apologies for all the questions, but if anyone could provide some insight I would much appreciate it.

Thanks in advance


===================== To manage your subscription to SPSSX-L, send a message to LISTSERV@LISTSERV.UGA.EDU (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

Back to: Top of message | Previous page | Main SPSSX-L page