SPSSX Discussion

jaccard measures in SPSS22

Classic

List

Threaded

3 messages Options

goodtoshare

jaccard measures in SPSS22

Dear all,

I'll be doing jaccard measures in Spss 22 to analyse the importance of brand
attributes. Any help or advice on data preparation and computing process is
much appreciated!

Data is gathered from questionnaire and it looks like the following: (to
simplify the case I'll show only 2
attributes (red and green) and 3 brands (brand A/B/C), while in fact there
are a lot more)

q1-Which brands do you think have the attribute red? （could be more than one
answer）--1. brand A, 2. brand B, 3. brand C

q2-Which brands do you think have the attribute green? （could be more than
one answer）--1. brand A, 2. brand B, 3. brand C

q3a/c/b-How would you score brand A/B/C respectively on a 1-5 point scale?

q4-Which of these three brands that you like most? (only one answer)--1.
brand A, 2. brand B, 3. brand C

q5-Which of these three brands that you will purchase again? (could be more
than one answer)--1. brand A, 2. brand B, 3. brand C
and the data might look like this:
<http://spssx-discussion.1045642.n5.nabble.com/file/t341380/%E9%98%BF%E9%87%8C%E6%97%BA%E6%97%BA%E5%9B%BE%E7%89%8720180101225808.jpg>

so what should I do on SPSS to calculate:
1. the importance of attributes red/ green towards questions like q3
2. the importance of attributes red/ green towards questions like q4
3. the importance of attributes red/ green towards questions like q5

--
Sent from: http://spssx-discussion.1045642.n5.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

Rich Ulrich

Re: jaccard measures in SPSS22

You start with asking about Jaccard measures (index, I assume), and I don't understand why.

I've pointed to that sort of information (X-not-Y, both/neither, Y-not-X) when trying to convince

an investigator that, despite a moderately impressive total N, that "information" is actually too

small to do useful analyses because all the cases look the same. You do not mention your N.

However: If you want to differentiate by (A, B) on a jaccard sort of basis, I guess you could score those

by computing, say GreenRedA = GreenA- RedA to measure the relevance of color for A, and similarly for

B; then look at the paired t-test for GreenRedA, GreenRedB. Each of these will be scaled scores, (-1,0,1).

For a large set of Brands, it would be a repeated measures ANOVA.

For your analyses with outcomes, I think you want to try to /write/ a statement of an assumed difference

or association with the outcome. What can you say that makes sense? - either, when the Brands are

equally favored, or when there are good and bad?

Rich Ulrich

From: SPSSX(r) Discussion <[hidden email]> on behalf of goodtoshare <[hidden email]>
Sent: Monday, January 1, 2018 10:03:00 AM
To: [hidden email]
Subject: jaccard measures in SPSS22

===================== 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

goodtoshare

Re: jaccard measures in SPSS22

Dear Rich,

Thanks for your reply and comments!

In fact I was able to contact the ex-colleague who did the jaccard
correlation in the first place and he was kind to share with me some of his
ideas. Hope the following info is useful to others as well :)

So the jaccard is chosen because of the binary data coming from questions
where more than one answers could be selected, and the key in data
preparation is to convert all data into 0/1. To do this, any one respondent
shall be divided into several cases based on the brands, and the 1-5 point
score shall be converted to 0/1 based on certain rules. (in the following
example, the score lower than 5 would be converted to 0, and only 5-point
would be 1)

To show this using my previous example and datamap, the data should actually
look like:

<http://spssx-discussion.1045642.n5.nabble.com/file/t341380/abc.png>

and then the usual jaccard measure could be used.

Hope this mini case would help others who are confused as I were. And thanks
again for Rich's kind reply!

--
Sent from: http://spssx-discussion.1045642.n5.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