INTRODUCTION

TO SAMPLING

1
 Purpose of Sampling
Why  sampling?  -­‐  to  study  the  whole  popula5on?    
A  major  reason  studying  samples  rather  than  the  whole  group  is  that  the  whole  group  is  so  large  that  
studying  it  is  not  feasible.  Example-­‐  college  students  in  CA.    If  we  can  study  the  whole  popula>on,  we  do  not  
need  to  go  through  the  sampling  procedures.  Much  research  is  based  on  samples  of  people.  
Representa5veness  -­‐  how  representa5ve  the  selected  cases  are?      
Then,  can  knowledge  gained  from  selected  cases  be  considered  knowledge  about  a  whole  group  of  
people?  The  answer  depends  on  whether  those  selected  cases  are  representa(ve  of  larger  group.    
Newsmagazine  ar>cles  about  public  opinion:  How  can  we  be  sure  that  the  results  reflect  the  public’s  
true  opinion,  in  other  words,  how  much  they  can  represent  views  of  all  Americans.  The  ul>mate  
purpose  of  sampling  is  to  get  accurate  representa(veness.  The  important  considera>on  about  
samples  is  how  representa>ve  they  are  of  the  popula>on  from  which  we  draw  them.    
Casual  vs.  scien5fic  sampling      
In  both  daily  life  and  prac>ce,  we  are  involved  in  sampling  decisions  -­‐  movies,  car  
purchases,  class  selec>ons,  etc;  to  get  feedbacks  about  service  sa>sfac>on  from  clients  –  
what  is  said  in  community  or  agency  mee>ng.    How  much  of  this  informa>on  is  
representa>ve?  The  informa>on  can  be  misleading  or  biased  -­‐  The  people  who  aLend  or  
are  the  most  vocal  at  a  mee>ng  may  be  the  most  sa>sfied  (or  most  dissa>sfied).  If  a  
sample  size  is  too  small,  informa>on  can  be  biased  as  well.    Scien>fic  sampling  is  
considerably  more  careful  and  systemic  than  casual,  everyday  sampling.  
In  research,  scien5fic  sampling  procedures  have  been  developed  so  that  we  can  
minimize  the  likelihood  that  samples  we  select  will  be  biased  or  too  small.      
2
 Purpose of Sampling (Cont.)
What  is  sampling?    The  process  of  drawing  a  subset  of  people  from  a  popula5on  
so  that  results  with  that  subset  may  be  generalized  to  the  popula5on.      See  below:  
 

     
 

3
  
Sampling Terminology
(Study/Target)  Popula5on    All  possible  cases  of  what  you  are  interested  in  
studying.  This  is  the  group  you  would  like  to  sample  from  because  this  is  the  group  you  
are  interested  in  generalizing  to.  A  sample  is  drawn  from  a  popula5on.  The  target  (study)  
popula5on  oKen  is  people  who  have  some  par5cular  characteris5cs  in  common,  such  as  
all  American,  all  eligible  voters,  all  school-­‐age  children,  and  so  on.  To  select  a  good  
sample,  we  need  to  clearly  define  the  popula5on  from  which  you  draw  the  sample.    
The  defini5on  of  popula5on  should  specify  four  items:  
 

   Content:  the  par>cular  characteris>cs  that  the  members  of    

the  popula>on  have  in  common  
 

       Unit:              unit  of  analysis  

 

       Extent:        spa>al  or  geographic  coverage  

 

       Time:          the  temporal  period  during  which  a  unit  must  

possess  the  appropriate  characteris>cs  to  qualify  for  
the  sample    
  4
 Sampling Terminology (Cont.)
 

Sampling  Frame  
   -­‐-­‐    A  lis5ng  of  all  element  (cases)  in  a  study  popula5on.  
   -­‐-­‐    In  many  studies,  we  draw  the  actual  sample  from  this  lis5ng.    

Examples:  lis5ngs  of  telephone  numbers,  or    customers  from  a  local  electric  u5lity  
tend  to  exclude  certain  groups,  such  as….????    telephone  books????    case  lists????  

The  adequacy  of  sampling  frame  is  crucial  in  determining  the  quality  
of  the  sample.  In  other  words,  the  degree  to  which  the  sampling  frame  includes  all  
members  of  the  popula5on  is  most  important.  Some  of  the  adequate  sampling  
frames  consist  of  lists  of  members  of  organiza5ons.  For  example,  membership  
rosters  of  NASW,  APA.  The  popula5on  consists  of  the  sampling  frame,  and  we  can  
make  legi5mate  generaliza5ons  only  about  the  sampling  frame.  Many  social  
workers,  for  example,  do  not  belong  to  NASW.  Thus,  a  sample  taken  from  the  NASW  
membership  roster  represents  only  NASW  members  and  not  all  social  workers.  It  is  
important  to  assess  carefully  who  the  list  includes  and  who  the  list  excludes.    
 
5
 Why Random Selection?
—  The word random refers to a process that
generates a mathematically random result, one
in which no humanly generated pattern exists.
—  Social work researchers usually try to select
their cases using a random procedure in order
to assure that no human bias exists in the
selection process. They hope that the
inferences they draw from their study will be
maximally generalizable, statistically accurate,
and useful. Using random procedures allows
the use of probability sampling methods.

6
 Sampling Techniques
 

Probability  Sampling  
   -­‐  based  on  probability  theory  
 -­‐  equal  probability  of  selec1on  can  ensure    
               representa>veness  
 -­‐  random  selec1on  
 -­‐  Can  es>mate  sampling  error  
Types  of  probability  sampling  
Simple  Random  Sampling  
Stra>fied  Sampling  
Cluster  Sampling  
    7

 
 Simple Random Sampling (SRS)
 

Ø  Applies  random  selec>on  using  random  numbers  

Ø  Generally  assumed  in  all  probability  sampling  applica>ons.    
Ø  Once  a  sampling  frame  has  been  established,  the  researcher  assigns  a  
single  number  to  each  member  in  the  list  without  skipping  any  number  in  
the  process.    
Ø  A  table  of  random  numbers  is  some>mes  used  to  select  element  for  the  
sample  
Ø  Many  computer  programs  can  generate  a  series  of  random  numbers.    
Ø  SRS  is  easy  to  accomplish  and  explain  to  others.    
Ø  SRS  oUen  is  imprac>cal,  especially  not  most  sta>s>cally  efficient  when  
dealing  with  large  scale  projects.      
Ø  Simple  vs.  not  most  efficient  &  not  good  representa>on  of  subgroups  
 

To  deal  with  these  issues  you  may  turn  to  other  sampling  techniques.    

 
  8

 
 Stratified Sampling
For  even  greater  precision,  and  to  ensure  adequate  
numbers  of  small  subgroups  (e.g.,  ethnic  minority  
groups)  in  the  sample.    
 

       (1)        Divide  the  sampling  frame  into    

                         homogeneous  subgroups  
                                                         ↓  
       (2)          Taking  a  SRS  in  each  subgroup  
 
You  will  be  able  to  represent  not  only  overall  
popula5on,  but  also  key  subgroups  of  popula5on,  
especially  small  minority  group  (age,  gender,  etc.).    
  9
 Cluster Sampling
You  may  have  to  sample  a  popula5on  that  is  spread  across  a  wide  
geographic  region.  Imagine  taking  a  SRS  of  all  the  residents  of  CA  to  
conduct  personal  interviews.  By  the  luck  of  the  draw,  you  will  wind  up  
with  respondents  who  come  from  all  over  the  state.  Your  interviewers  
are  going  to  have  a  lot  of  traveling  to  do.    
Cluster  random  sampling  addresses  the  problem.  Use  the  follow  steps:  
 
(1)  Divide  a  popula>on  into  clusters    
                 (usually  along  geographic  boundaries)  or  some  popula5ons  are  
already  grouped  into  clusters  (e.g.,  churches,  schools)  
       ↓  
 (2)  randomly  select  clusters  
       ↓    
 (3)  measure  all  elements  within  sampled  clusters.    
   
We  carry  out  cluster  sampling  mainly  because  of  efficiency  of  
administra5on,  and  to  reduce  the  need  to  even  larger  samples.      
 
  10
  
Nonprobability Sampling
We  can  use  probability  sampling  techniques  only  when  we  can  have  a  
sampling  frame.  Many  5mes  in  social  work  research,  it  is  impossible  to  
develop  a  sampling  frame  of  a  popula5on.  For  example,  undocumented  
immigrants  are,  by  defini5on,  not  listed  anywhere  
Perhaps  you  just  want  to  evaluate  a  program  in  a  your  agency  and  don’t  care  
whether  the  program  would  work  with  other  people  in  other  places  and  at  
other  5mes.    
Researchers  oKen  must  rely  on  their  judgment  to  purposively  select  a  sample  
that  seems  representa5ve  or  that  seems  to  fit  the  purpose  of  the  inquiry.  
Some5mes  they  simply  rely  on  available  subjects  
All  of  the  sampling  techniques  that  do  not  involve  the  use  of  probability  
methods  are  called  nonprobability  sampling  (do  not  involve  random  
selec5on).  Generally,  probability  sampling  is  the  preferred  method.  
However,  in  some  circumstances  in  applied  social  research,  it  is  not  feasible,  
prac5cal,  or  theore5cally  sensible  to  use  random  sampling.    
When  you  use  nonprobability  sampling,  it  is  important  to  be  cau5ous  and  
mindful  of  the  risks  inherent  in  it.     11
 
Convenience and Purposive Sampling
 

Convenience  (availability)  Sampling:  

           -­‐-­‐relying  on  available  subjects  
           -­‐-­‐inexpensive  and  uses  ready  access  to  a  certain  popula5on  
           -­‐-­‐you  can  ask  for  volunteers  and  provide  incen5ves  to  them  
           -­‐-­‐weakness  is  that  you  have  no  evidence  of  representa5veness  
           -­‐-­‐you  must  jus5fy  that  this  is  the  most  feasible  method  
 
Purposive  Sampling:  
                 -­‐-­‐based  on  researcher’s  judgment  or  prior  knowledge  
               -­‐-­‐suppose  that  you  want  to  examine  effect  of  support  group  
among  female,  adult  children,  of  aged  adults–  you  would  
purposively  choose  aged  adult  caregivers,  who  are  female,  
adult  children,  and  never  been  in  a  support  group.    
 
 

12
 Quota and Snowball Sampling
 
Quota  Sampling  
 
You  rely  on  available  subjects  but  strive  to  representa5ves  by  construc5ng  matrix  represen5ng  one  or  more  
characteris5cs  (gender,  age,  educa5on,  religion,  race,  etc.),  and  then  collec5ng  data  from  people  who  had  all  
the  characteris5cs  in  a  given  parameter.    
 
You  con5nue  sampling  for  each  cell  un5l  you  get  the  desired  number  and  then  stop.  
 
If  you  already  have  40  women  for  your  sample,  but  not  the  60  men,  you  would  con5nue  to  sample  men.  If  
eligible  women  respondents  come  along,  you  would  not  sample  them  because  you  have  already  met  your  
quota.      
 
Although  using  quotas  may  improve  representa5veness,  it  is  s5ll  nonprobability  sampling  and  rely  on  
available  subjects  -­‐  oKen  depending  on  who  comes  along  when.      
 
Snowball  Sampling  
 
You  begin  by  iden5fying  people  who  meet  the  criteria  for  inclusion  in  your  study.    
 
Then  you  ask  them  to  recommend  others  they  know  who  also  meet  the  criteria.    
 
Subjects  are  accumulated  gradually  in  a  snowball  fashion.  For  example,  if  you  are  studying  
undocumented  immigrants,  you  are  not  likely  to  find  good  lists  of  immigrants  within  a  specific  
geographic  area.  However,  if  you  iden5fy  one  or  two,  you  may  find  that  they  know  who  the  other  
illegal  immigrants  in  the  area  and  how  you  can  find  them.  Useful  in  inves5ga5ng  in  sensi5ve  topics,  
such  as  child  abuse  or  drug  use,  where  the  perpetrators  or  the  vic5ms  might  hesitate  to  iden5fy  
themselves  if  approached  by  a  stranger,  such  as  a  researcher,  but  might  be  open  to  an  approach  by  
someone  who  they  know  shares  their  experiences  or  deviant  status.    
 
Although  probability  sampling  is  less  risky  than  nonprobability  sampling  from  the  standpoint  of  
generalizing  accurately  to  a  popula5on,  some  good  studies  use  nonprobability  sampling.  Rather  than  
think  you  must  always  use  probability  sampling,  you  should  understand  when  certain  sampling  
techniques  are  more  appropriate  than  others,  the  func5ons  of  each  techniques,  how  feasibility  
constraints  bear  on  the  choice  of  a  sample  procedures,  and  the  risks  inherent  when  samples  are  too  
small  or  vulnerable  to  bias.    
 
13
 Extreme Case, Sequential, and
Theoretical Sampling
—  Extreme Case – select cases based on
their unusualness or difficulty of
finding
—  Sequential – select cases based on
some preset order of selection
—  Theoretical – select cases according
to theory

14
 Determining Sample Size
• Should  consider  the  number  of  variables,  the  amount  of  
sampling  errors,  popula5on  homogeneity,  sampling  frac5on  
(the  number  of  element  in  the  sample  rela5ve  to  the  
number  of  elements  in  the  popula5on),  and  sampling  
techniques  (more  complex  sampling  techniques  requires  
bigger  sample  sizes).    
• OKen  determined  by  mul5plying  the  number  of  variables  
by  the  minimum  number  of  cases  per  variables  required  by  
the  appropriate  sta5s5cal  procedure    
• Sta5s5cal  power  analysis:  how  large  a  sample  needs  to  be  
in  order  for  researchers  to  have  an  adequate  probability  of  
obtaining  sta5s5cally  significant  findings.    
 
  15
Sensitivity to Diversity in Sampling
Regardless  of  whether  you  are  using  probability  or  nonprobability  
sampling  techniques,  you  should  be  careful  to  avoid  biases:  
 

Ø  Gender  bias  –  do  not  generalize  to  both  genders  when  one  gender  is  not  
adequately  represented  in  the  research  sample.    
 

Ø  Use  cultural  sensi5vity  in  all  phases  of  research,  including  sampling.    
 
Ø  Carefully  examine  whether  certain  minority  groups  are  inadequately  
represented  in  the  sample,  or  unwarranted  generaliza5ons  are  made  to  the  
en5re  popula5on.    

Ø  Employ  local  community  members  as  research  staff    

Ø  Providing  transporta5on  and  child  care  or  data  collec5on  at  home    
Ø  Choose  a  sensi5ve  and  accessible  sebng  -­‐  may  not  want  to  travel  to  strange  
area,  convenient  and  safe  sebng  

Ø  Use  and  train  culturally  competent  interviewers  &  bilingual  staff   16