Announcement

Collapse
No announcement yet.
X
Collapse
  • Page of 1
  • Filter
  • Time
  • Show
Clear All
new posts
  • #1

    Modeling a skewed categorical (ordinal) variable

    Hi Statalist,

    The response scale of my dependent variable is structured as "Not at all" "To a small extent" "To a moderate extent" and "To a great extent." The survey question asks "Have your organization used social media for the following purposes?" 1) Newsletter, 2) Important announcements, 3) Customer service, 4) Events, 5) General community engagement, and 6) Recruitment.
    My current strategy is to build a composite measure of social media use (eg., egen rowmean) where "Not at all" translates to 1, "Moderate extent" to 2, ..., and run it as an OLS model.
    But the distribution is highly positively skewed where more than half responded to "Not at all" for most categories. I wonder if I should employ a different strategy (eg., multinomial logit, logit, count models).

    My question is...
    1. Is it OK to construct a composite score scale and run an OLS?
    2. Is there a way I can model it as a composite measure rather than modeling the 6 subdimensions separately?
    3. Any other suggestions I would highly appreciate.

    Here is an example of my data. Thanks.

    Code:
    * Example generated by -dataex-. To install: ssc install dataex
clear
input str20(sm_newsletter sm_annoucements sm_customer sm_events sm_engagement sm_recruit)
"Not at all"           "To a great extent"    "To a great extent"    "To a great extent"    "To a moderate extent" "To a small extent"   
"Not at all"           "To a small extent"    "Not at all"           "To a small extent"    "Not at all"           "To a small extent"   
"To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"   
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a moderate extent" "To a small extent"    "To a great extent"    "To a moderate extent" "To a small extent"   
"Not at all"           "To a great extent"    "To a moderate extent" "To a great extent"    "To a moderate extent" "To a great extent"   
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"To a great extent"    "To a great extent"    "To a great extent"    "NA"                   "NA"                   "NA"                  
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"   
"To a moderate extent" "To a great extent"    "To a moderate extent" "To a moderate extent" "To a moderate extent" "Not at all"          
"To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"   
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"Not at all"           "To a moderate extent" "To a small extent"    "To a moderate extent" "To a moderate extent" "Not at all"          
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"To a small extent"    "To a moderate extent" "To a moderate extent" "To a moderate extent" "To a moderate extent" "To a moderate extent"
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"To a moderate extent" "To a great extent"    "To a moderate extent" "To a great extent"    "To a moderate extent" "Not at all"          
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"To a moderate extent" "To a great extent"    "To a great extent"    "To a moderate extent" "To a moderate extent" "To a small extent"   
"To a small extent"    "To a small extent"    "To a small extent"    "To a small extent"    "To a small extent"    "To a small extent"   
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"To a great extent"    "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"To a moderate extent" "To a moderate extent" "To a moderate extent" "NA"                   "To a moderate extent" "To a small extent"   
"Not at all"           "Not at all"           "To a great extent"    "To a great extent"    "Not at all"           "To a small extent"   
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a great extent"    "Not at all"           "To a great extent"    "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a moderate extent" "To a moderate extent" "To a moderate extent" "To a moderate extent" "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"To a great extent"    "To a great extent"    "To a moderate extent" "To a great extent"    "To a great extent"    "To a moderate extent"
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a great extent"    "To a great extent"    "To a moderate extent" "To a moderate extent" "To a great extent"   
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"   
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "To a small extent"    "To a small extent"    "To a small extent"   
"To a great extent"    "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a moderate extent" "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a small extent"    "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a small extent"    "To a small extent"    "To a small extent"    "Not at all"           "Not at all"          
"To a small extent"    "To a moderate extent" "To a moderate extent" "To a moderate extent" "To a moderate extent" "To a small extent"   
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"To a moderate extent" "To a moderate extent" "To a moderate extent" "To a moderate extent" "To a moderate extent" "To a moderate extent"
"Not at all"           "To a small extent"    "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "To a small extent"    "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a small extent"    "Not at all"           "To a small extent"    "To a small extent"    "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"    "To a great extent"    "To a small extent"   
"Not at all"           "Not at all"           "To a small extent"    "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"To a small extent"    "To a small extent"    "To a moderate extent" "To a small extent"    "To a moderate extent" "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "To a small extent"    "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "To a small extent"    "Not at all"          
"To a moderate extent" "To a moderate extent" "To a moderate extent" "To a moderate extent" "To a moderate extent" "To a moderate extent"
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a small extent"    "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a small extent"    "To a small extent"    "Not at all"           "Not at all"           "Not at all"          
"NA"                   "NA"                   "NA"                   "NA"                   "NA"                   "NA"                  
"Not at all"           "Not at all"           "Not at all"           "To a moderate extent" "To a moderate extent" "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a moderate extent" "Not at all"           "Not at all"           "To a small extent"    "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "NA"                   "Not at all"          
"Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"           "Not at all"          
"Not at all"           "To a great extent"    "To a small extent"    "Not at all"           "Not at all"           "Not at all"          
end
    Tags: None
  • #2
    Originally posted by Yong-Jin Ahn View Post
    the distribution is highly positively skewed where more than half responded to "Not at all" for most categories. I wonder if I should employ a different strategy (eg., multinomial logit, logit, count models).
    It probably depends upon what you think is going on behind an organization's not using social media, but how about something along the following lines?

    (For the sake of argument here, "NA" is considered equivalent to "Not at all" whenever social media are used for at least one other purpose.)

    .ÿ
    .ÿversionÿ17.0

    .ÿ
    .ÿclearÿ*

    .ÿ
    .ÿ//ÿseedem
    .ÿsetÿseedÿ501858466

    .ÿ
    .ÿquietlyÿinputÿstr20(sm_newsletterÿsm_annoucementsÿsm_customerÿsm_eventsÿsm_engagementÿsm_recruit)

    .ÿ
    .ÿ*
    .ÿ*ÿBeginÿhere
    .ÿ*
    .ÿlabelÿdefineÿScoresÿ0ÿÿ"Notÿatÿall"ÿ1ÿ"Toÿaÿsmallÿextent"ÿ2ÿ"Toÿaÿmoderateÿextent"ÿ3ÿ"Toÿaÿgreatÿextent"ÿ.mÿ"NA"

    .ÿforeachÿvarÿofÿvarlistÿsm_*ÿ{
    ÿÿ2.ÿÿÿÿÿÿÿÿÿlocalÿnewÿ=ÿsubstr("`var'",ÿ4,ÿ3)
    ÿÿ3.ÿÿÿÿÿÿÿÿÿencodeÿ`var',ÿgenerate(`new')ÿlabel(Scores)ÿnoextend
    ÿÿ4.ÿÿÿÿÿÿÿÿÿdropÿ`var'
    ÿÿ5.ÿ}

    .ÿ
    .ÿquietlyÿegenÿbyteÿscoÿ=ÿrowtotal(new-rec),ÿmissing

    .ÿtabulateÿsco,ÿmissing

    ÿÿÿÿÿÿÿÿscoÿ|ÿÿÿÿÿÿFreq.ÿÿÿÿÿPercentÿÿÿÿÿÿÿÿCum.
    ------------+-----------------------------------
    ÿÿÿÿÿÿÿÿÿÿ0ÿ|ÿÿÿÿÿÿÿÿÿ45ÿÿÿÿÿÿÿ45.00ÿÿÿÿÿÿÿ45.00
    ÿÿÿÿÿÿÿÿÿÿ1ÿ|ÿÿÿÿÿÿÿÿÿÿ7ÿÿÿÿÿÿÿÿ7.00ÿÿÿÿÿÿÿ52.00
    ÿÿÿÿÿÿÿÿÿÿ2ÿ|ÿÿÿÿÿÿÿÿÿÿ2ÿÿÿÿÿÿÿÿ2.00ÿÿÿÿÿÿÿ54.00
    ÿÿÿÿÿÿÿÿÿÿ3ÿ|ÿÿÿÿÿÿÿÿÿÿ7ÿÿÿÿÿÿÿÿ7.00ÿÿÿÿÿÿÿ61.00
    ÿÿÿÿÿÿÿÿÿÿ4ÿ|ÿÿÿÿÿÿÿÿÿÿ2ÿÿÿÿÿÿÿÿ2.00ÿÿÿÿÿÿÿ63.00
    ÿÿÿÿÿÿÿÿÿÿ6ÿ|ÿÿÿÿÿÿÿÿÿÿ2ÿÿÿÿÿÿÿÿ2.00ÿÿÿÿÿÿÿ65.00
    ÿÿÿÿÿÿÿÿÿÿ7ÿ|ÿÿÿÿÿÿÿÿÿÿ3ÿÿÿÿÿÿÿÿ3.00ÿÿÿÿÿÿÿ68.00
    ÿÿÿÿÿÿÿÿÿÿ8ÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿÿÿÿÿÿ1.00ÿÿÿÿÿÿÿ69.00
    ÿÿÿÿÿÿÿÿÿÿ9ÿ|ÿÿÿÿÿÿÿÿÿÿ3ÿÿÿÿÿÿÿÿ3.00ÿÿÿÿÿÿÿ72.00
    ÿÿÿÿÿÿÿÿÿ10ÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿÿÿÿÿÿ1.00ÿÿÿÿÿÿÿ73.00
    ÿÿÿÿÿÿÿÿÿ11ÿ|ÿÿÿÿÿÿÿÿÿÿ2ÿÿÿÿÿÿÿÿ2.00ÿÿÿÿÿÿÿ75.00
    ÿÿÿÿÿÿÿÿÿ12ÿ|ÿÿÿÿÿÿÿÿÿÿ4ÿÿÿÿÿÿÿÿ4.00ÿÿÿÿÿÿÿ79.00
    ÿÿÿÿÿÿÿÿÿ13ÿ|ÿÿÿÿÿÿÿÿÿÿ3ÿÿÿÿÿÿÿÿ3.00ÿÿÿÿÿÿÿ82.00
    ÿÿÿÿÿÿÿÿÿ16ÿ|ÿÿÿÿÿÿÿÿÿÿ2ÿÿÿÿÿÿÿÿ2.00ÿÿÿÿÿÿÿ84.00
    ÿÿÿÿÿÿÿÿÿ18ÿ|ÿÿÿÿÿÿÿÿÿÿ4ÿÿÿÿÿÿÿÿ4.00ÿÿÿÿÿÿÿ88.00
    ÿÿÿÿÿÿÿÿÿÿ.ÿ|ÿÿÿÿÿÿÿÿÿ12ÿÿÿÿÿÿÿ12.00ÿÿÿÿÿÿ100.00
    ------------+-----------------------------------
    ÿÿÿÿÿÿTotalÿ|ÿÿÿÿÿÿÿÿ100ÿÿÿÿÿÿ100.00

    .ÿ
    .ÿ//ÿMaybeÿsomethingÿlikeÿthis,ÿforÿexample
    .ÿquietlyÿfmm:ÿ(pointmassÿsco,ÿvalue(0))ÿ(regressÿsco)ÿ//ÿThere'sÿnoÿ-nolog-ÿoption

    .ÿfmm

    FiniteÿmixtureÿmodelÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿ=ÿ88
    Logÿlikelihoodÿ=ÿ-194.38867

    ------------------------------------------------------------------------------
    ÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿCoefficientÿÿStd.ÿerr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿconf.ÿinterval]
    -------------+----------------------------------------------------------------
    1.Classÿÿÿÿÿÿ|ÿÿ(baseÿoutcome)
    -------------+----------------------------------------------------------------
    2.Classÿÿÿÿÿÿ|
    ÿÿÿÿÿÿÿ_consÿ|ÿÿÿ.0135486ÿÿÿÿ.220009ÿÿÿÿÿ0.06ÿÿÿ0.951ÿÿÿÿ-.4176612ÿÿÿÿ.4447584
    ------------------------------------------------------------------------------

    Class:ÿÿÿÿ2ÿÿÿÿÿÿ
    Response:ÿscoÿÿÿÿ
    Model:ÿÿÿÿregress

    ------------------------------------------------------------------------------
    ÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿCoefficientÿÿStd.ÿerr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿconf.ÿinterval]
    -------------+----------------------------------------------------------------
    scoÿÿÿÿÿÿÿÿÿÿ|
    ÿÿÿÿÿÿÿ_consÿ|ÿÿÿÿ7.47211ÿÿÿ.8670931ÿÿÿÿÿ8.62ÿÿÿ0.000ÿÿÿÿÿ5.772638ÿÿÿÿ9.171581
    -------------+----------------------------------------------------------------
    ÿÿÿvar(e.sco)|ÿÿÿ31.50774ÿÿÿ6.726878ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ20.73415ÿÿÿÿ47.87933
    ------------------------------------------------------------------------------

    .ÿ
    .ÿ//ÿAnotherÿpossibility
    .ÿgenerateÿdoubleÿpreÿ=ÿcond(!sco,ÿrnormal(1,ÿ1),ÿrnormal(0,ÿ1))ÿ//ÿInÿyourÿcase,ÿuseÿaÿrealÿvariable

    .ÿziologitÿsco,ÿinflate(c.pre)ÿnolog

    Zero-inflatedÿorderedÿlogitÿregressionÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿ=ÿÿÿÿÿ88
    ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿWaldÿchi2(1)ÿÿ=ÿÿÿ3.44
    Logÿlikelihoodÿ=ÿ-158.1106ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿProbÿ>ÿchi2ÿÿÿ=ÿ0.0637

    ------------------------------------------------------------------------------
    ÿÿÿÿÿÿÿÿÿscoÿ|ÿCoefficientÿÿStd.ÿerr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿconf.ÿinterval]
    -------------+----------------------------------------------------------------
    inflateÿÿÿÿÿÿ|
    ÿÿÿÿÿÿÿÿÿpreÿ|ÿÿ-1.128619ÿÿÿ.6086288ÿÿÿÿ-1.85ÿÿÿ0.064ÿÿÿÿ-2.321509ÿÿÿÿ.0642717
    ÿÿÿÿÿÿÿ_consÿ|ÿÿÿ1.025976ÿÿÿÿ1.15139ÿÿÿÿÿ0.89ÿÿÿ0.373ÿÿÿÿ-1.230707ÿÿÿÿ3.282659
    -------------+----------------------------------------------------------------
    ÿÿÿÿÿÿÿ/cut1ÿ|ÿÿ-1.829202ÿÿÿ1.725556ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-5.21123ÿÿÿÿ1.552827
    ÿÿÿÿÿÿÿ/cut2ÿ|ÿÿ-.9513933ÿÿÿ.8901259ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-2.696008ÿÿÿÿ.7932214
    ÿÿÿÿÿÿÿ/cut3ÿ|ÿÿ-.7598344ÿÿÿ.7884534ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-2.305175ÿÿÿÿ.7855059
    ÿÿÿÿÿÿÿ/cut4ÿ|ÿÿ-.1645562ÿÿÿ.5795954ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-1.300542ÿÿÿÿÿÿ.97143
    ÿÿÿÿÿÿÿ/cut5ÿ|ÿÿ-.0038758ÿÿÿ.5440936ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-1.07028ÿÿÿÿ1.062528
    ÿÿÿÿÿÿÿ/cut6ÿ|ÿÿÿ.1567545ÿÿÿ.5153896ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-.8533906ÿÿÿÿ1.166899
    ÿÿÿÿÿÿÿ/cut7ÿ|ÿÿÿ.4022363ÿÿÿ.4828723ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-.544176ÿÿÿÿ1.348649
    ÿÿÿÿÿÿÿ/cut8ÿ|ÿÿÿ.4864237ÿÿÿ.4745335ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-.4436447ÿÿÿÿ1.416492
    ÿÿÿÿÿÿÿ/cut9ÿ|ÿÿÿ.7509234ÿÿÿ.4565901ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-.1439768ÿÿÿÿ1.645824
    ÿÿÿÿÿÿ/cut10ÿ|ÿÿÿÿ.844531ÿÿÿ.4530092ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-.0433508ÿÿÿÿ1.732413
    ÿÿÿÿÿÿ/cut11ÿ|ÿÿÿ1.043351ÿÿÿ.4498129ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.1617344ÿÿÿÿ1.924968
    ÿÿÿÿÿÿ/cut12ÿ|ÿÿÿ1.513986ÿÿÿÿ.464308ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.603959ÿÿÿÿ2.424013
    ÿÿÿÿÿÿ/cut13ÿ|ÿÿÿÿ1.99023ÿÿÿ.5088264ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.9929484ÿÿÿÿ2.987511
    ÿÿÿÿÿÿ/cut14ÿ|ÿÿÿ2.440242ÿÿÿÿ.579072ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ1.305282ÿÿÿÿ3.575203
    ------------------------------------------------------------------------------

    .ÿ
    .ÿexit

    endÿofÿdo-file


    .

    I'm not that familiar with these modeling approaches and their subtleties (assumptions, gotchas etc.), but they might be worth looking into further in your case.
    • 1 like

    Comment

    • #3
      Thanks, Joseph. Can you elaborate a bit on the logic/rationale for your suggestions?

      Comment

      • #4
        Originally posted by Yong-Jin Ahn View Post
        Hi Statalist,

        The response scale of my dependent variable is structured as "Not at all" "To a small extent" "To a moderate extent" and "To a great extent." The survey question asks "Have your organization used social media for the following purposes?" 1) Newsletter, 2) Important announcements, 3) Customer service, 4) Events, 5) General community engagement, and 6) Recruitment.
        My current strategy is to build a composite measure of social media use (eg., egen rowmean) where "Not at all" translates to 1, "Moderate extent" to 2, ..., and run it as an OLS model.
        But the distribution is highly positively skewed where more than half responded to "Not at all" for most categories. I wonder if I should employ a different strategy (eg., multinomial logit, logit, count models).

        My question is...
        1. Is it OK to construct a composite score scale and run an OLS?
        2. Is there a way I can model it as a composite measure rather than modeling the 6 subdimensions separately?
        3. Any other suggestions I would highly appreciate.

        ...
        Answering from a psychometrics perspective: If you want to create a composite scale, you should first ensure that the 6 items can be scaled that way. In a lot of use cases, researchers will start with a validated scale. If you didn't do this, then you should consider this next time. Naturally, if there isn't a measure of the thing you're trying to measure, that means more work for your team, so there are tradeoffs.

        In this case, my intuition is that your 6 items might not form an acceptable composite measure. Normally, I deal with scales that measure the level of depressive symptoms, the level of physical disability, how good someone's health-related quality of life is, etc. What sort of construct does your scale measure? Something like the overall extent of social media use? Or is it just measuring 6 different aspects of social media use? It seems more like the latter to me. I apologize if this is cryptic, this is hard to convey briefly over the internet. However, again, my instinct is that a) I don't know if those 6 items would form an adequate unitary scale, b) if you want to treat them as one composite scale, then ideally you need to validate that scale first, and c) you would certainly be justified in running six separate (ordered logistic, ordered probit, maybe generalized ordered logit) regressions and presenting the results.

        In healthcare, it's very common to have skewed composite scales and to run OLS or a relative of OLS (e.g. hierarchical linear model or GEE for repeated measures). The skewedness by itself is not a big issue, statistically. Consider the common saying that "all models are wrong, some are useful" (usually attributed to George Box). There are more advanced techniques that can be applied (e.g. health economists often recommend a gamma GLM, log link for healthcare cost data, which are skewed, but I think Poisson is sometimes useful as an alternative), but these can be harder to use and interpret. However, OLS is still useful, even though it may be a wrong model.
        Be aware that it can be very hard to answer a question without sample data. You can use the dataex command for this. Type help dataex at the command line.

        When presenting code or results, please use the code delimiters format them. Use the # button on the formatting toolbar, between the " (double quote) and <> buttons.

        Comment

        • #5
          Originally posted by Yong-Jin Ahn View Post
          Can you elaborate a bit on the logic/rationale for your suggestions?
          From your description of your data's distribution, those commands sounded as if they might be apt.

          You didn't mention anything about your research question, but I assumed that it has something to do with assessing association between characteristics of an organization and its proclivity to use social media for various of its needs. This usually entails something like fitting regression models, and those approaches seem to accommodate the peculiarity of your data.

          You might want to look further into them (check motivating examples that are used illustratively in their help files and user manual entries, for example) in order to see whether either is a good match to your needs.

          My question is...
          1. Is it OK to construct a composite score scale and run an OLS?
          Initially it seems as if a simple sumscore might be defensible, but given that many of your organizations that use social media for at least one purpose nevertheless chose to answer "NA" instead of "Not at all" to other needs suggests that these purposes won't neatly lump themselves together into a single dimension, and so militate against a sumscore.

          Weiwen beat me to the punch here, but I'm inclined to agree with him that some or perhaps all of these questionnaire items need to be handled individually.

          Comment

          • #6
            Thanks Weiwen and Joseph!

            Comment

            Previous Next
            Working...
            X