Coping with individual response patterns

[John F Hall]

I need some advice on how to deal with widely differing individual response patterns to (mostly) 0-10 scales used to measure satisfaction with life and life domains, happiness, anxiety and other constructs in the SPSS files for the following surveys:

 

SSRC Survey Unit Quality of Life in Britain surveys (1971-1975)

ONS Well-being survey, Unrestricted Access Teaching Data Set (April 2011)

ONS Well-being survey (merged data set April – August 2011)

British Social Attitudes (2008 and 2013)

European Social Survey (Wave 6, 2012)

 

My problem is how to treat responses when some are spread out across the whole scale and others are clustered.  (See sample data below).  Can I assume everyone is using the scales in the same way and that the ratings are valid, or do I need to build in some statistical controls to take account of the variations in individual response patterns? 

 

There’s a full account in my draft working paper (32pp):

http://surveyresearch.weebly.com/uploads/2/9/9/8/2998485/patterns_of_individual_responses_on_0-10_scales_used_to_measure_satisfaction_and_other_well-being_constructs_a_comparison_of_usage_in_different_surveys.pdf

 

All comments and suggestions gratefully received.

 

John F Hall (Mr)

[Retired academic survey researcher]

 

Email:   [hidden email] 

Website: www.surveyresearch.weebly.com

SPSS start page:  www.surveyresearch.weebly.com/1-survey-analysis-workshop

 

In the data extracts below, cases highlighted in red have used few points spread across the whole scale: cases highlighted in blue have also used few points, but they are clustered (usually towards the top of the scale).

 

 

R uses three points only

 

zeros ones twos threes fours fives sixes sevens eights nines tens

 

   3    0    0     0      0     7     0     0      0      0   52

   2    0    0     0      0     2     0     0      0      0   36

   7    0    0     0      0    18     0     0      0      0   13

 

   0    0    0     0      0     0     0     1      1      0   57

   0    0    0     0      0     0     0     1      5      0   34

   0    0    0     0      0     0     0     1      7      0   33

 

R uses four points only

 

zeros ones twos threes fours fives sixes sevens eights nines tens

 

   3    0    0     0      0    12     0     0     11      0   36

   3    0    0     0      0     5     0     0      0      1   30

   6    0    0     0      0     3     0     0      1      0   31

   5    0    0     0      0     6     0     0      1      0   29

   1    0    0     0      0     5     0     0      9      0   46

   9    0    1     0      0    16     0     0      0      0   15

   1    0    0     0      0     8     0     0      8      0   24

 

   0    0    0     0      0     1     0     0      5      1   33

  0    0    0     0      0     7     0     1      1      0   31

   0    0    0     0      1     0     0     0      4     14   22

   0    0    0     0      0     0     0     1      7      6   27

   0    0    0     0      0     0     0     1      1     12   27

   0    0    0     0      0     9     0     1      0      1   51

   0    0    0     0      0     1     1     0      0      2   56

   0    0    0     0      0     0     0     1      6      4   29

   0    0    0     0      0     6     2     0     23      0    5

 

R uses five points only

 

zeroes ones twos threes fours fives sixes sevens eights nines tens

 

   4    0    0     0      0     2     0     5      3      0   48

   8    0    0     0      0     7     0     0      1      1   45

   2    0    0     4      0     0     3     0      7      0   24

  14    1    0     0      0    13     0     3      0      0   10

  16    0    0     0      0     8     0     1      1      0   15

  16    0    1     0      0    13     0     0      2      0    8

   2    0    0     0      0    15     0     3      2      0   40

 

 

   0    0    0     0      0     0     2     4      6      8   21

   0    0    0     0      0     7     0     3     10      5   16

   0    0    0     0      0     0     9    12     11     14   10

   0    0    0     0      0     0     1     2      2      1   35

   0    0    0     0      0     1     0     1     12     12   35

   0    0    0     0      0    10     3    11     13      0    4

   0    0    0     0      0     5     0     1     11      4   41

   0    0    0     0      0     1     0     1      2      1   36

   0    0    0     0      0     0     8    10     18     12   14

   0    0    0     0      0     1     0     2      4     14   20

   0    0    0     0      0     0     5    14     36      5    2

 

 

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

[Kirill Orlov]

-->Can I assume everyone is using the scales in the same way and that the ratings are valid, or do I need to build in some statistical controls to take account of the variations in individual response patterns?

I think it is very hard  to advice, to give any "universal" or "best" recipe. You say in all the surveys the scales used are same (10-point likert (or 11-level??). Were they really same in their graphical layout? (I.e. numbers everywhere or boxes everywhere...)?

The red data show focal effects on "round" numbers 0, 5, 10 and a bit on 8, while blue data show diffuse ceiling effect. Various reasons could predispose to both.

Were instructions or situational subliminal prompts for respondents really similar in the surveys? In what respect respondents (sample's) social background differed there (age, edication level...)?

(10-point likerts are bad, especially to unexperienced/young respondents). I would vote for using 7-point, at max.)

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