ASA Connect

 View Only

  • 1.  Comparing two set of responses

    Posted 01-25-2017 12:51
    HI ,
     I need a suggestion about which statistical test to use. 
    We have two sets of response. Response one is a survey data based on 0-10 <g class="gr_ gr_225 gr-alert gr_spell gr_run_anim ContextualSpelling ins-del multiReplace" id="225" data-gr-id="225">likert</g> scale data.  And the second one <g class="gr_ gr_341 gr-alert gr_gramm gr_run_anim Style multiReplace" id="341" data-gr-id="341">is  a</g> survey based on 0-3 <g class="gr_ gr_310 gr-alert gr_spell gr_run_anim ContextualSpelling ins-del multiReplace" id="310" data-gr-id="310">likert</g> scale data. I am thinking of comparing Pearson coefficient skewness of the two responses. Any suggestion would be appreciated.

    Thanks



    ------------------------------
    Amit Saha

    ------------------------------


  • 2.  RE: Comparing two set of responses

    Posted 01-26-2017 05:03
    Amit

    Let's back up a little bit. What's your research objective here?

    Blaise

    ------------------------------
    Blaise Egan
    Lead Data Scientist
    British Telecommunications PLC
    ------------------------------

    Original Message


  • 3.  RE: Comparing two set of responses

    Posted 01-26-2017 07:39
    Blaise, Our research objective is to compare the two surveys methods. We believe the first survey results are skewed towards the right side and <g class="gr_ gr_257 gr-alert gr_gramm gr_run_anim Grammar only-ins replaceWithoutSep" id="257" data-gr-id="257">second</g> survey is somewhat better than the first. We want to statistically see if our belief is correct or not.

    Thanks 

    ------------------------------
    Amit Saha

    ------------------------------

    Original Message


  • 4.  RE: Comparing two set of responses

    Posted 01-28-2017 21:34
    Forgive me if this is a dumb question, but are you administering both survey methods to every subject in your survey?

    ------------------------------
    Eric Siegel, MS
    Research Associate
    Department of Biostatistics
    Univ. Arkansas Medical Sciences
    ------------------------------

    Original Message


  • 5.  RE: Comparing two set of responses

    Posted 01-30-2017 07:33
    Hi Eric,
    We are doing the two survey over <g class="gr_ gr_103 gr-alert gr_gramm gr_run_anim Grammar only-ins replaceWithoutSep" id="103" data-gr-id="103">same</g> population over time. Our earlier survey response on 10 points scale. The second response survey we started using is based on 4 points scale. We want to statistically analyze the two survey and determine which has <g class="gr_ gr_526 gr-alert gr_gramm gr_run_anim Grammar only-ins doubleReplace replaceWithoutSep" id="526" data-gr-id="526">better</g> range of response.
    Thank
    <g class="gr_ gr_538 gr-alert gr_spell gr_run_anim ContextualSpelling ins-del multiReplace" id="538" data-gr-id="538">amit</g>

    ------------------------------
    Amit Saha

    ------------------------------

    Original Message


  • 6.  RE: Comparing two set of responses

    Posted 02-01-2017 00:07

    Same population, but separated in time, you say? Interesting. How far apart in time are the two surveys administered?

    Confidentiality Notice: This e-mail message, including any attachments, is for the sole use of the intended recipient(s) and may contain confidential and privileged information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply e-mail and destroy all copies of the original message.



    Original Message


  • 7.  RE: Comparing two set of responses

    Posted 01-26-2017 21:27
    Unfortunately, what many call a "Likert Scale" does not meet the definition created by Likert in the 1930's.  A Likert scale is one where a difference of one unit at one end of the scale is equivalent to a difference of one unit throughout the scale.  In this way, the numbers generated can be treated as measurements of some specific quality.   Most "Likert Scales" are just arbitrary sets of numbers (1-10, 1-4, etc.), and treating them the same way we treat temperatures from a thermometer, for instance, is a mistake.

    I would go back to the condition being "measured" by the so-called Likert scales and find regions of values for each scale that represent qualitative, but ordered, characteristics of the problem being examined.  I'd construct three of four characteristics, like "deterioration", "no change", "moderate improvement", "excellent improvement", put these categories into a contingency table and use a powerful test for ordered differences, like the Cochran-Armitage test, to determine if there is a difference between "treatments" or whatever is being compared.  If you feel the circumstances that led to the use of two difference scales are sufficient to raise questions about the comparability of the two studies, you might want to use a log-linear model on two contingency tables.

    David Salsburg




    Original Message


  • 8.  RE: Comparing two set of responses

    Posted 02-02-2017 09:44
    Hi Amit,

    Now remember, I haven't seen your research so I may be way off here...but...

    If you sum the scores for each respondent on both of your scales, then you can basically take care of the Likert/ordinal problem because now you have "scores" so you can assume continuous variable.

    Then you could compare the two measures with a Pearson's correlation just fine, because Pearson's doesn't care about the scale of the two variables. That is a nice thing about correlation.

    I am also assuming these are the same people at both times. There may be a problem with the two tests being given at different times, but if the tests were on the same people you most likely would only have added error caused by the method or process, not much error added for the instrument or scale.

    Hope this helps!
    Elaine

    ------------------------------
    Elaine Eisenbeisz
    Owner and Principal Statistician
    Omega Statistics
    ------------------------------

    Original Message


Related Content