DISCOVERING

 STATISTICS  USING  SPSS  

Chapter 5: The beast of bias

Smart Alex’s Solutions

Task 1
Using   the   [Link]   data   from   Chapter   4,   check   the   assumptions   of   normality   and  
homogeneity  of  variance  for  the  two  films  (ignore  gender):  are  the  assumptions  met?  
 
The  output  you  should  get  is  reproduced  below  (I  used  the  explore  function  described  in  Chapter  5).  

 
The  skewness  statistics  gives  rise  to  a  z-­‐score  of  −0.378/0.512  =  –0.74  for  Bridget  Jones’s  Diary,  and  
0.04/0.512   =   0.08   for   Memento.   These   show   no   significant   skewness.   For   kurtosis   these   values   are  
−0.254/0.992  =  –0.26  for  Bridget  Jones’s  Diary,  and  –1.024/0.992  =  –1.03,  which  again  are  both  non-­‐
significant.  
The   Q-­‐Q   plots   confirm   these   findings:   for   both   films   the   expected   quantile   points   are   close   to  
those  that  would  be  expected  from  a  normal  distribution  (i.e.  the  dots  fall  close  to  the  diagonal  line).  
The   K-­‐S   tests   show   no   significant   deviation   from   normality   for   both   films.   We   could   report   that  
arousal  scores  for  Bridget  Jones’s  Diary,  D(20)  =  0.13,  ns,  and  Memento,  D(20)  =  0.10,  ns,  were  both  
not   significantly   different   from   a   normal   distribution.   Therefore   we   can   assume   normality   in   the  
sample  data.  
In  terms  of  homogeneity  of  variance,  Levene’s  test  shows  that  the  variances  of  arousal  for  the  two  
films  were  not  significantly  different,  F(1,  38)  =  1.90.  
 

PROFESSOR  ANDY  P  FIELD     1  

DISCOVERING  STATISTICS  USING  SPSS  

   

PROFESSOR  ANDY  P  FIELD     2  

DISCOVERING  STATISTICS  USING  SPSS  

 
 

Task 2
The  file  [Link]  contains  data  regarding  students’  performance  on  an  SPSS  exam.  Four  
variables   were   measured:   exam   (first-­‐year   SPSS   exam   scores   as   a   percentage),   computer  
(measure  of  computer  literacy  in  percent),  lecture  (percentage  of  SPSS  lectures  attended)  and  
numeracy  (a  measure  of  numerical  ability  out  of  15).  There  is  a  variable  called  uni  indicating  
whether   the   student   attended   Sussex   University   (where   I   work)   or   Duncetown   University.  
Compute  and  interpret  descriptive  statistics  for  exam,  computer,  lecture,  and  numeracy  for  the  
sample  as  a  whole.  
To   see   the   distribution   of   the   variables,   we   can   use   the   frequencies   command,   which   we   came   across  
in   the   previous   section   (see   Section   5.3.2).   Use   this   dialog   box   and   place   all   four   variables   (exam,  
computer,  lecture  and  numeracy)  in  the  Variable(s)  box.  Then  click  on    to  select  the  Statistics  
dialog   box   and   select   some   measures   of   central   tendency   (mean,   mode,   median),   measures   of  
variability   (range,   standard   deviation,   variance,   quartile   splits)   and   measures   of   shape   (kurtosis   and  
skewness).  Also  click  on    to  access  the  Charts  dialog  box  and  select  a  frequency  distribution  of  
scores   with   a   normal   curve.   Return   to   the   main  dialog  box  by  clicking  on     and   once   in   the   main  
dialog  box,  click  on    to  run  the  analysis.  
The   output   shows   the   table   of   descriptive   statistics   for   the   four   variables   in   this   example.   From  
this   table,   we   can   see   that,   on   average,   students   attended   nearly   60%   of   lectures,   obtained   58%   in  
their  SPSS  exam,  scored  only  51%  on  the  computer  literacy  test,  and  only  5  out  of  15  on  the  numeracy  
test.  In  addition,  the  standard  deviation  for  computer  literacy  was  relatively  small  compared  to  that  of  
the  percentage  of  lectures  attended  and  exam  scores.  These  latter  two  variables  had  several  modes  
(multimodal).    
The   output   provides   tabulated   frequency   distributions   of   each   variable   (not   reproduced   here).  
These  tables  list  each  score  and  the  number  of  times  that  it  is  found  within  the  data  set.  In  addition,  
each   frequency   value   is   expressed   as   a   percentage   of   the   sample   (in   this   case   the   frequencies   and  
percentages  are  the  same  because  the  sample  size  was  100).  Also,  the  cumulative  percentage  is  given,  
which  tells  us  how  many  cases  (as  a  percentage)  fell  below  a  certain  score.  So,  for  example,  we  can  

PROFESSOR  ANDY  P  FIELD     3  

DISCOVERING  STATISTICS  USING  SPSS  

see  that  66%  of  numeracy  scores  were  5  or  less,  74%  were  6  or  less,  and  so  on.  Looking  in  the  other  
direction,  we  can  work  out  that  only  8%  (100  − 92%)  got  scores  greater  than  8.  

 
Output  

Finally,   we   are   given   histograms   of   each   variable   with   the   normal   distribution   overlaid.   These  
graphs  show  us  several  things.  The  exam  scores  are  very  interesting  because  this  distribution  is  quite  
clearly   not   normal;   in   fact,   it   looks   suspiciously   bimodal   (there   are   two   peaks,   indicative   of   two  
modes).   This   observation   corresponds   with   the   earlier   information   from   the   table   of   descriptive  
statistics.   It   looks   as   though   computer   literacy   is   fairly   normally   distributed   (a   few   people   are   very  
good   with   computers   and   a   few   are   very   bad,   but   the   majority   of   people   have   a   similar   degree   of  
knowledge)   as   is   the   lecture   attendance.   Finally,   the   numeracy   test   has   produced   very   positively  
skewed   data   (the   majority   of   people   did   very   badly   on   this   test   and   only   a   few   did   well).   This  
corresponds  to  what  the  skewness  statistic  indicated.  
Descriptive   statistics   and   histograms   are   a   good   way   of   getting   an   instant   picture   of   the  
distribution  of  your  data.  This  snapshot  can  be  very  useful:  for  example,  the  bimodal  distribution  of  
SPSS  exam  scores  instantly  indicates  a  trend  that  students  are  typically  either  very  good  at  statistics  or  
struggle  with  it  (there  are  relatively  few  who  fall  in  between  these  extremes).  Intuitively,  this  finding  
fits  with  the  nature  of  the  subject:  statistics  is  very  easy  once  everything  falls  into  place,  but  before  
that  enlightenment  occurs  it  all  seems  hopelessly  difficult!    

   

PROFESSOR  ANDY  P  FIELD     4  

DISCOVERING  STATISTICS  USING  SPSS  

   

Figure:  Histograms  of  the  SPSS  exam  data  

Task 3
Calculate  and  interpret  the  z-­‐scores  for  skewness  for  all  variables.  
 
For  the  SPSS  exam  scores,  the  z-­‐score  of  skewness  is  −0.107/0.241  =  −0.44.  For  numeracy,  the  z-­‐score  
of  skewness  is  0.961/0.241  =  3.99.  For  computer  literacy,  the  z-­‐score  of  skewness  is  −0.174/0.241  =  
−0.72.  For  lectures  attended,  the  z-­‐score  of  skewness  is  −0.422/0.241  =  −1.75.  It  is  pretty  clear  then  
that  the  numeracy  scores  are  significantly  positively  skewed  (p  <  .05)  because  the  z-­‐score  is  greater  
than   1.96,   indicating   a   pile-­‐up   of   scores   on   the   left   of   the   distribution   (so   most   students   got   low  
scores).  For  the  other  three  variables,  the  skewness  is  non-­‐significant,  p  <  .05,  because  the  values  lie  
between  −1.96  and  1.96.  

Task 4
Calculate  and  interpret  the  z-­‐scores  for  kurtosis  for  all  variables.  
 
For   SPSS   exam   scores,   the   z-­‐score   of   kurtosis   is   −1.105/0.478   =   −2.31,   which   is   significant,   p   <   .05,  
because  it  lies  outside  −1.96  and  1.96.  
For  computer  literacy,  the  z-­‐score  of  kurtosis  is  0.364/0.478  =  0.76,  which  is  non-­‐significant,  p  <  .05,  
because  it  lies  between  −1.96  and  1.96.  
For  lectures  attended,  the  z-­‐score  of  kurtosis  is  −0.179/0.478  =  −0.37,  which  is  non-­‐significant,  p  <  .05,  
because  it  lies  between  −1.96  and  1.96.  
For  numeracy,  the  z-­‐score  of  kurtosis  is  0.946/0.478  =  1.98,  which  is  significant,  p  <  .05,  because  it  lies  
outside  −1.96  and  1.96.  

Task 5
Use  the  split  file  command  to  look  at  and  interpret  the  descriptive  statistics  for  numeracy  and  
exam.  
If  we  want  to  obtain  separate  descriptive  statistics  for  each  of  the  universities,  we  can  split  the  file,  
and   then   proceed   using   the   frequencies   command.   To   split   the   file,   select     or   click   on  
.   In   the   resulting   dialog   box   select   the   option   Organize   output   by   groups.   Once   this   option   is  
selected,   the   Groups   Based   on   box   will   activate.   Select   the   variable   containing   the   group   codes   by  

PROFESSOR  ANDY  P  FIELD     5  

DISCOVERING  STATISTICS  USING  SPSS  

which  you  wish  to  repeat  the  analysis  (in  this  example  select  Uni),  and  drag  it  to  the  box  or  click  on  
.   By   default,   SPSS   will   sort   the   file   by   these   groups   (i.e.   it   will   list   one   category   followed   by   the  
other   in   the   data   editor   window).   Once   you   have   split   the   file,   use   the   frequencies   command.   Let’s  
request  statistics  for  only  numeracy  and  exam  scores  for  the  time  being.  

 
Dialog  box  for  the  split  file  command  

Duncetown  University   Sussex  University  

   

SPSS  output  

The  SPSS  output  is  split  into  two  sections:  first  the  results  for  students  at  Duncetown  University,  
then   the   results   for   those   attending   Sussex   University.   The   output   shows   the   two   main   summary  
tables.  From  these  tables  it  is  clear  that  Sussex  students  scored  higher  on  both  their  SPSS  exam  and  
the  numeracy  test  than  their  Duncetown  counterparts.  In  fact,  looking  at  the  means  reveals  that,  on  
average,  Sussex  students  scored  an  amazing  36%  more  on  the  SPSS  exam  than  Duncetown  students,  
and  had  higher  numeracy  scores  too  (what  can  I  say,  my  students  are  the  best).  
 

PROFESSOR  ANDY  P  FIELD     6  

DISCOVERING  STATISTICS  USING  SPSS  

SPSS  Exam  Mark  

   

Numeracy  

   

Distributions  of  exam  and  numeracy  scores  for  Duncetown  University  and  Sussex  University  students  

The  figure  shows  the  histograms  of  these  variables  split  according  to  the  university  attended.  The  
first   interesting   thing   to   note   is   that   for   exam   marks,   the   distributions   are   both   fairly   normal.   This  
seems  odd  because  the  overall  distribution  was  bimodal.  However,  it  starts  to  make  sense  when  you  
consider  that  for  Duncetown  the  distribution  is  centred  around  a  mark  of  about  40%,  but  for  Sussex  
the  distribution  is  centred  around  a  mark  of  about  76%.  This  illustrates  how  important  it  is  to  look  at  
distributions   within   groups.   If   we   were   interested   in   comparing   Duncetown   to   Sussex   it   wouldn’t  
matter   that   overall   the   distribution   of   scores   was   bimodal;   all   that’s   important   is   that   each   group  
comes  from  a  normal  distribution,  and  in  this  case  it  appears  to  be  true.  When  the  two  samples  are  
combined,  these  two  normal  distributions  create  a  bimodal  one  (one  of  the  modes  being  around  the  
centre  of  the  Duncetown  distribution,  and  the  other  being  around  the  centre  of  the  Sussex  data!).  For  
numeracy  scores,  the  distribution  is  slightly  positively  skewed  (there  is  a  larger  concentration  at  the  
lower  end  of  scores)  in  both  the  Duncetown  and  Sussex  groups.  Therefore,  the  overall  positive  skew  
observed   before   is   due   to   the   mixture   of   universities.   When   you   have   finished   with   the   split   file  
command,  remember  to  switch  it  off  (otherwise  SPSS  will  carry  on  doing  every  analysis  on  each  group  
separately).   To   switch   this   function   off,   return   to   the   Split   File   dialog   box   and   select   Analyze   all   cases,  
do  not  create  groups.  

Task 6
Repeat  Task  5  but  for  the  computer  literacy  and  percentage  of  lectures  attended.  

PROFESSOR  ANDY  P  FIELD     7  

 DISCOVERING  STATISTICS  USING  SPSS  

 
The  SPSS  output  is  split  into  two  sections:  first,  the  results  for  students  at  Duncetown  University,  then  
the   results   for   those   attending   Sussex   University.   From   these   tables   it   is   clear   that   Sussex   and  
Duncetown   students   scored   similarly   on   computer   literacy   (both   means   are   very   similar).   Sussex  
students  attended  slightly  more  lectures  (63.27%)  than  their  Duncetown  counterparts  (56.26%).  
The  histograms  are  also  split  according  to  the  university  attended.  All  of  the  distributions  look  fairly  
normal.   The   only   exception   is   the   computer   literacy   scores   for   the   Sussex   students.   This   is   a  fairly   flat  
distribution  apart  from  a  huge  peak  between  50  and  60%.  It’s  slightly  heavy-­‐tailed  (right  at  the  very  
ends  of  the  curve  the  bars  come  above  the  line)  and  very  pointy.  This  suggests  positive  kurtosis.  If  you  
examine   the   values   of   kurtosis   you   will   find   that   there   is   significant   (p   <   .05)   positive   kurtosis:  
1.38/0.662  =  2.08,  which  falls  outside  of  −1.96  and  1.96.    
 
Duncetown  University   Sussex  University  

   

 
Computer  Literacy  

   

Percentage  of  Lectures  Attended  

PROFESSOR  ANDY  P  FIELD     8  

DISCOVERING  STATISTICS  USING  SPSS  

 
 
 

Task 7
Conduct  and  interpret  a  K-­‐S  test  for  numeracy  and  exam.  
 
The   Kolmogorov–Smirnov   (K-­‐S)   test   can   be   accessed   through   the   explore   command   (
).   First,   enter   any   variables   of   interest   in   the   box   labelled  
Dependent  List  by  highlighting  them  on  the  left-­‐hand  side  and  transferring  them  by  clicking  on   .  For  
this  example,  just  select  the  exam  scores  and  numeracy  scores.  It  is  also  possible  to  select  a  factor  (or  
grouping   variable)   by   which   to   split   the   output   (so   if   you   select   Uni   and   transfer   it   to   the   box   labelled  
Factor   List,   SPSS   will   produce   exploratory   analysis   for   each   group   —   a   bit   like   the   split   file   command).  
If   you   click   on     a   dialog   box   appears,   but   the   default   option   is   fine   (it   will   produce   means,  
standard  deviations  and  so  on).  The  more  interesting  option  for  our  current  purposes  is  accessed  by  
clicking   on   .   In   this   dialog   box   select   the   option   ,   and   this   will   produce  
both   the   K-­‐S   test   and   some   graphs   called   normal   Q-­‐Q   plots.   Click   on     to   return   to   the   main  
dialog  box  and  then  click  on    to  run  the  analysis.  

PROFESSOR  ANDY  P  FIELD     9  

DISCOVERING  STATISTICS  USING  SPSS  

 
Dialog  boxes  for  the  explore  command  

The  first  table  produced  by  SPSS  contains  descriptive  statistics  (mean,  etc.)  and  should  have  the  
same   values   as   the   tables   obtained   using   the   frequencies   procedure.   The   important   table   is   that   of  
the  K-­‐S  test,  which  includes  the  test  statistic  itself,  the  degrees  of  freedom  (which  should  equal  the  
sample  size)  and  the  significance  value  of  this  test.  Remember  that  a  significant  value  (Sig.  less  than  
.05)   indicates   a   deviation   from   normality.   For   both   numeracy   and   SPSS   exam   scores,   the   K-­‐S   test   is  
highly  significant,  indicating  that  both  distributions  are  not  normal.  This  result  is  likely  to  reflect  the  
bimodal   distribution   found   for   exam   scores,   and   the   positively   skewed   distribution   observed   in   the  
numeracy   scores.   However,   these   tests   confirm   that   these   deviations   were   significant.   (But   bear   in  
mind  that  the  sample  is  fairly  big.)  

 
SPSS  Output  

We  can  report  the  results  in  the  SPSS  output  in  the  following  way:  
• The  percentages  on  the  SPSS  exam,  D(100)  =  0.10,  p  <  .05,  and  the  numeracy  scores,  D(100)  
=  0.15,  p  <  .001,  were  both  significantly  non-­‐normal.  

PROFESSOR  ANDY  P  FIELD     10  

DISCOVERING  STATISTICS  USING  SPSS  

As   a   final   point,   bear   in   mind   that   when   we   looked   at   the   exam   scores   for   separate   groups,   the  
distributions  seemed  quite  normal;  now  if  we’d  asked  for  separate  tests  for  the  two  universities  (by  
placing  Uni  in  the  box  labelled  Factor  List)  the  K-­‐S  test  might  not  have  been  significant.  In  fact  if  you  
try   this   out,   you’ll   find   that   the   percentages   on   the   SPSS   exam   are   indeed   normal   within   the   two  
groups  (the  values  in  the  Sig.  column  are  greater  than  .05).  This  is  important  because  if  our  analysis  
involves  comparing  groups,  then  what’s  important  is  not  the  overall  distribution  but  the  distribution  
in  each  group.    

Tests of Normality
a
Kolmogorov-Smirnov Shapiro-Wilk
University Statistic df Sig. Statistic df Sig.
Percentage on Duncetown University .106 50 .200* .972 50 .283
SPSS exam Sussex University .073 50 .200* .984 50 .715
Numeracy Duncetown University .183 50 .000 .941 50 .015
Sussex University .155 50 .004 .932 50 .007
*. This is a lower bound of the true significance.
a. Lilliefors Significance Correction
 
SPSS  0utput    

Normal Q-Q Plot of SPSS Exam Scores Normal Q-Q Plot of Numeracy
3
3

2 2

1
1

0
Expected Normal

Expected Normal

-1

-2 -1

-3
-2
0 20 40 60 80 100 120
-2 0 2 4 6 8 10 12 14 16

Observed Value
  Observed Value
 

Normal  Q-­‐Q  plots  of  numeracy  and  SPSS  exam  scores

SPSS  also  produces  a  normal  Q-­‐Q  plots.  In  both  of  the  variables  analysed  we  already  know  that  the  
data  are  not  normal,  and  these  plots  confirm  this  observation  because  the  dots  deviate  substantially  
from   the   line.   It   is   noteworthy   that   the   deviation   is   greater   for   the   numeracy   scores,   and   this   is  
consistent  with  the  higher  significance  value  of  this  variable  on  the  K-­‐S  test.  

Task 8
Conduct  and  interpret  a  Levene’s  test  for  numeracy  and  exam.  
We   can   get   Levene’s   test   using   the   Explore   menu:   .  
Transfer   the   SPSS   exam   scores   and   the   numeracy   scores   from   the   list   on   the   left-­‐hand   side   to   the   box  
labelled   Dependent   List   by   clicking   on   the     next   to   this   box,   and   because   we   want   to   split   the  
output  by  the  grouping  variable  to  compare  the  variances,  select  the  variable   Uni  and  transfer  it  to  
the   box   labelled   Factor   List   by   clicking   on   the   appropriate   .   Then   click   on     to   open   another  
dialog  box.  To  get  Levene’s  test  we  need  to  select  one  of  the  options  where  it  says  Spread  vs.  level  
with  Levene’s  test.  If  you  select    Levene’s  test  is  carried  out  on  the  raw  data  (a  good  
place  to  start).  When  you’ve  finished  with  this  dialog  box  click  on    to  return  to  the  main  Explore  
dialog  box  and  then  click  on    to  run  the  analysis.  

PROFESSOR  ANDY  P  FIELD     11  

DISCOVERING  STATISTICS  USING  SPSS  

 
Exploring  groups  of  data  and  obtaining  Levene’s  test  

Levene’s  test  is  non-­‐significant  for  the  SPSS  exam  scores  (values  in  the  [Link]  are  more  than  
.05),   indicating   that   the   variances   are   not   significantly   different   (i.e.   they   are   similar   and   the  
homogeneity  of  variance  assumption  is  tenable).  However,  for  the  numeracy  scores,  Levene’s  test  is  
significant  (values  in  the  Sig.  column  are  less  than  .05),  indicating  that  the  variances  are  significantly  
different  (i.e.,  they  are  not  the  same  and  the  homogeneity  of  variance  assumption  has  been  violated).  
We  can  also  calculate  the  variance  ratio.  For  SPSS  exam  scores  the  variance  ratio  is  158.48/104.14  =  
1.52   and   for   numeracy   scores   the   value   is   9.43/4.27   =   2.21.   Our   group   sizes   are   50   and   we’re  
comparing   two   variances   so   the   critical   value   is   (from   the   table   in   the   additional   material)  
approximately   1.67.   These   ratios   concur   with   Levene’s   test:   variances   are   significantly   different   for  
numeracy  scores  (2.21  is  bigger  than  1.67)  but  not  for  SPSS  exam  scores  (1.52  is  smaller  than  1.67).  
We  could  report  these  findings  as  follows:  
• For  the  percentage  on  the  SPSS  exam,  the  variances  were  equal  for  Duncetown  and  Sussex  
University   students,   F(1,   98)   =   2.58,   ns,   but   for   numeracy   scores   the   variances   were  
significantly  different  in  the  two  groups,  F(1,  98)  =  7.37,  p  <  .01.  

 
 
SPSS  output    

Task 9
Transform  the  numeracy  scores  (which  are  positively  skewed)  using  one  of  the  transformations  
described  in  this  chapter.  Do  the  data  become  normal?  
 

PROFESSOR  ANDY  P  FIELD     12  

DISCOVERING  STATISTICS  USING  SPSS  

These   are   the   original   histogram   and   those   of   the   transformed   scores   (I’ve   included   all   three  
transformations  discussed  in  the  chapter):  

   

   

None  of  these  histograms  appear  to  be  normal.  Below  is  the  table  of  results  from  the  K–S  test,  all  of  
which   are   significant.   The   only   conclusion   is   that   although   the   square   root   transformation   does   the  
best  job  of  normalizing  the  data,  none  of  these  transformations  actually  works!  

 
 

PROFESSOR  ANDY  P  FIELD     13  

DISCOVERING  STATISTICS  USING  SPSS  

Task 10
Use  the  explore  command  to  see  what  effect  a  natural  log  transformation  would  have  on  the  
four  variables  measured  in  [Link].  
The  completed  dialog  boxes  should  look  like  this:  

 
The   SPSS   output   below   shows   Levene’s   test   on   the   log-­‐transformed   scores.   Compare   this   table   to   the  
one  in  the  book  (which  was  conducted  on  the  untransformed  SPSS  exam  scores  and  numeracy).  To  
recap  the  book  chapter,  for  the  untransformed  scores  Levene’s  test  was  non-­‐significant  for  the  SPSS  
exam   scores   (the   value   in   the   column   labelled   Sig.   was   .111,     more   than   .05)   indicating   that   the  

PROFESSOR  ANDY  P  FIELD     14  

DISCOVERING  STATISTICS  USING  SPSS  

variances   were   not   significantly   different   (i.e.,   the   homogeneity   of   variance   assumption   is   tenable).  
However,  for  the  numeracy  scores,  Levene’s  test  was  significant  (the  value  in  the  column  labelled  Sig.  
was  .008,  less  than  .05)  indicating  that  the  variances  were  significantly  different  (i.e.  the  homogeneity  
of  variance  assumption  was  violated).  
For   the   log-­‐transformed   scores   (below),   the   problem   has   been   reversed:   Levene’s   test   is   now  
significant   for   the   SPSS   exam   scores   (values   in   the   column   labelled   Sig.   are   less   than   .05)   but   is   no  
longer   significant   for   the   numeracy   scores   (values   in   the   column   labelled  Sig.   are   more   than   .05).   This  
reiterates   my   point   from   the   book   chapter   that   transformations   are   often   not   a   magic   solution   to  
problems  in  the  data!    

 
 
 

PROFESSOR  ANDY  P  FIELD     15