Discovery	
  or	
  Displacement?	
  
A	
  Major	
  Longitudinal	
  Study	
  of	
  the	
  Effect	
  
of	
  Web-­‐Scale	
  Discovery	
  Services	
  on	
  
Online	
  (Journal)	
  Usage	
  
SCELC	
  Colloquium	
  
March	
  5,	
  2014	
  
	
  
Michael	
  Levine-­‐Clark,	
  University	
  of	
  Denver	
  
John	
  McDonald,	
  University	
  of	
  Southern	
  California	
  
Jason	
  Price,	
  SCELC	
  ConsorNum	
  
“…a steep increase in full text downloads
and link resolver click‐throughs suggests
Summon had a dramatic impact on user
behavior and the use of library collections
during this time period.”
The Impact ofWeb-scale Discovery on the Use of a Library
Collection
Doug Way (2010) http://scholarworks.gvsu.edu/
library_sp/9/
h"p://www.oclc.org/partnerships/econtent/solu4ons.en.html	
  
Vendor	
  marke5ng	
  
Does	
  implementa4on	
  of	
  a	
  discovery	
  
service	
  impact	
  usage	
  of	
  publisher-­‐
hosted	
  journal	
  content?	
  
What	
  did	
  we	
  measure?	
  
•  Whether	
  there	
  is	
  an	
  
effect	
  
•  NOT	
  why	
  that	
  effect	
  
exists	
  (that’s	
  a	
  future	
  
study!)	
  
•  “Society	
  will	
  need	
  to	
  shed	
  some	
  of	
  its	
  obsession	
  
for	
  causality	
  in	
  exchange	
  for	
  simple	
  correla5ons:	
  
not	
  knowing	
  why,	
  but	
  only	
  what”	
  	
  
•  Cukier	
  &	
  Mayer-­‐Schonberger.	
  2013.	
  Big	
  data:	
  A	
  revolu4on	
  that	
  will	
  
transform	
  how	
  we	
  live,	
  work,	
  and	
  think.	
  
Data	
  collec5on	
  
•  List	
  of	
  libraries	
  with	
  discovery	
  services	
  
> Searched	
  on	
  lib-­‐web-­‐cats	
  
•  Surveyed	
  Libraries	
  
> Discovery	
  service	
  Implemented	
  
> ImplementaNon	
  Date	
  (month/year)	
  
> Search	
  box	
  locaNon	
  
> MarkeNng	
  effort	
  
•  149	
  Libraries	
  Gave	
  Approval	
  
> 33	
  libraries	
  selected	
  for	
  this	
  phase	
  
> 6	
  for	
  each	
  of	
  the	
  4	
  major	
  discovery	
  services	
  and	
  a	
  
group	
  of	
  9	
  libraries	
  with	
  no	
  service	
  
Dataset	
  •  33	
  Libraries	
  
– 28	
  US,	
  2	
  CA,	
  1	
  each	
  from	
  UK,	
  AUS,	
  NZ	
  
– WorldCat	
  book	
  holdings	
  
>  Average:	
  1,114,193	
  ;	
  Range:	
  ~300k	
  to	
  ~2.6mil	
  
•  ImplementaNon	
  dates	
  (Discovery	
  Libraries):	
  	
  
>  2010	
  (3),	
  2011	
  (19),	
  2012	
  (2)	
  
•  6	
  Publishers	
  
•  9,206	
  Journals	
  
•  163,545	
  Usable	
  ObservaNons	
  
	
  
Methodology	
  
Compared	
  COUNTER	
  JR1	
  total	
  full	
  text	
  arNcle	
  views	
  for	
  the	
  
12	
  months	
  before	
  vs	
  12	
  months	
  aeer	
  implementaNon	
  date	
  
	
  
	
  
	
  
	
  	
  	
  
	
  
June	
  2010	
  Start	
  
ImplementaNon	
  
May	
  2011	
  
May	
  2012	
  
End	
  
Year	
  1	
   Year	
  2	
  
Included	
  implementaNon	
  month	
  in	
  Year	
  1	
  to	
  ensure	
  that	
  
both	
  periods	
  included	
  an	
  enNre	
  academic	
  year	
  
Examine	
  Data	
  for	
  Outliers	
  
Observa5ons	
  by	
  Library	
  &	
  Service	
  
Observa5ons	
  by	
  Publisher	
  
Average	
  Usage	
  Change	
  By	
  Discovery	
  &	
  Publisher	
  
Analyzing	
  Usage	
  Change:	
  %	
  vs	
  Total	
  	
  
Use	
  12	
  
months	
  
before	
  
Use	
  12	
  
months	
  
aRer	
  
%	
  Change	
  
Total	
  
Change	
  
Journal	
  A	
   500	
   600	
   20%	
   100	
  
Journal	
  B	
   5	
   15	
   200%	
   10	
  
Which	
  is	
  the	
  beier	
  measure?	
  	
  
Is	
  it	
  the	
  same	
  for	
  publisher-­‐	
  &	
  journal-­‐level	
  data?	
  	
  	
  
Reducing	
  varia5on	
  due	
  to	
  ins5tu5on	
  size	
  
Currently	
  converNng	
  to	
  change	
  per	
  FTE	
  	
  
	
  
Values	
  are	
  shown	
  as	
  x	
  1,000	
  to	
  bring	
  the	
  change	
  
metric	
  back	
  per	
  journal-­‐library	
  combinaNon	
  to	
  a	
  
minimum	
  of	
  0.1	
  
	
  
2013	
  JISC	
  Discovery	
  study	
  took	
  a	
  similar	
  approach	
  
Average	
  Usage	
  Change	
  By	
  Discovery	
  &	
  Publisher	
  
Per	
  Journal	
  &	
  
Per	
  10,000	
  FTE	
  
Full	
  Model	
  
Including	
  Discovery	
  Service,	
  Publisher,	
  and	
  Library	
  
Including	
  Discovery	
  Service,	
  Publisher,	
  and	
  Library	
  
Does the effect of discovery service differ across libraries?
Library	
  10-­‐15	
   Library	
  16-­‐21	
   Library	
  22-­‐27	
   Library	
  28-­‐33	
  Library	
  1-­‐9	
  
Nested	
  ANOVA	
  Model	
  	
  
[all	
  three	
  factors	
  –	
  preliminary	
  results]	
  
Does	
  usage	
  change	
  vary	
  across	
  libraries?	
  	
  
Institution (sorted by Mean Change)
Does	
  usage	
  change	
  vary	
  across	
  publishers?	
  	
  
Publisher (sorted by Mean Change)
Does	
  usage	
  change	
  vary	
  across	
  discovery	
  services?	
  
Publisher
Does	
  the	
  effect	
  of	
  discovery	
  service	
  differ	
  across	
  publishers?	
  	
  
Results	
  
Can	
  we	
  detect	
  differences	
  between	
  Discovery	
  
Services,	
  Publishers,	
  and/or	
  Libraries	
  and/or	
  
their	
  interac4ons?	
  	
  
•  Library	
  –	
  Yes	
  
•  Publisher	
  –	
  No	
  
•  Discovery	
  Service	
  –	
  Yes	
  
•  DifferenNal	
  discovery	
  service	
  effect	
  by	
  
publisher	
  –	
  Yes	
  
Next	
  Steps	
  
•  Design	
  &	
  test	
  for	
  effects	
  of:	
  	
  
–  Aggregator	
  full	
  text	
  availability	
  
–  Publisher	
  Size	
  
–  Journal	
  Subject	
  
–  Overall	
  usage	
  trends	
  (Requires	
  Disc	
  Srvc	
  ‘control’)	
  
–  ConfiguraNon	
  opNons	
  in	
  Discovery	
  services	
  
•  Expand	
  pool	
  of	
  libraries	
  
•  Perhaps	
  explore	
  WHY	
  
	
  
	
  	
  
Sharing	
  Data	
  
•  With	
  par5cipa5ng	
  libraries	
  
–  Customized	
  reports	
  for	
  each	
  library	
  
•  With	
  par5cipa5ng	
  publishers	
  
–  Customized	
  reports	
  for	
  each	
  publisher	
  
–  PresentaNons	
  as	
  requested	
  
•  With	
  discovery	
  vendors	
  
–  PresentaNons	
  as	
  requested	
  
•  In	
  publica5ons	
  and	
  presenta5ons	
  
–  Maintaining	
  anonymity	
  of	
  data	
  
	
  	
  
Doing	
  “Resarch”,	
  SCELC	
  Style!	
  
•  Why	
  SCELC?	
  
•  SCELC	
  Funding	
  
–  staNsNcs	
  consultant	
  	
  
–  research	
  &	
  wriNng	
  retreats	
  
–  See	
  hip://bit.ly/1dNMDL3	
  for	
  more	
  detail	
  
•  SCELC	
  libraries	
  encouraged	
  to	
  parNcipate	
  in	
  next	
  round	
  
–  Survey:	
  hip://bit.ly/DSparNcipaNon	
  
	
  
	
  	
  

Discovery impact scelc colloquium 2014mar05

  • 1.
    Discovery  or  Displacement?   A  Major  Longitudinal  Study  of  the  Effect   of  Web-­‐Scale  Discovery  Services  on   Online  (Journal)  Usage   SCELC  Colloquium   March  5,  2014     Michael  Levine-­‐Clark,  University  of  Denver   John  McDonald,  University  of  Southern  California   Jason  Price,  SCELC  ConsorNum  
  • 2.
    “…a steep increasein full text downloads and link resolver click‐throughs suggests Summon had a dramatic impact on user behavior and the use of library collections during this time period.” The Impact ofWeb-scale Discovery on the Use of a Library Collection Doug Way (2010) http://scholarworks.gvsu.edu/ library_sp/9/
  • 3.
  • 4.
    Does  implementa4on  of  a  discovery   service  impact  usage  of  publisher-­‐ hosted  journal  content?  
  • 5.
    What  did  we  measure?   •  Whether  there  is  an   effect   •  NOT  why  that  effect   exists  (that’s  a  future   study!)  
  • 6.
    •  “Society  will  need  to  shed  some  of  its  obsession   for  causality  in  exchange  for  simple  correla5ons:   not  knowing  why,  but  only  what”     •  Cukier  &  Mayer-­‐Schonberger.  2013.  Big  data:  A  revolu4on  that  will   transform  how  we  live,  work,  and  think.  
  • 9.
    Data  collec5on   • List  of  libraries  with  discovery  services   > Searched  on  lib-­‐web-­‐cats   •  Surveyed  Libraries   > Discovery  service  Implemented   > ImplementaNon  Date  (month/year)   > Search  box  locaNon   > MarkeNng  effort   •  149  Libraries  Gave  Approval   > 33  libraries  selected  for  this  phase   > 6  for  each  of  the  4  major  discovery  services  and  a   group  of  9  libraries  with  no  service  
  • 10.
    Dataset  •  33  Libraries   – 28  US,  2  CA,  1  each  from  UK,  AUS,  NZ   – WorldCat  book  holdings   >  Average:  1,114,193  ;  Range:  ~300k  to  ~2.6mil   •  ImplementaNon  dates  (Discovery  Libraries):     >  2010  (3),  2011  (19),  2012  (2)   •  6  Publishers   •  9,206  Journals   •  163,545  Usable  ObservaNons    
  • 11.
    Methodology   Compared  COUNTER  JR1  total  full  text  arNcle  views  for  the   12  months  before  vs  12  months  aeer  implementaNon  date                 June  2010  Start   ImplementaNon   May  2011   May  2012   End   Year  1   Year  2   Included  implementaNon  month  in  Year  1  to  ensure  that   both  periods  included  an  enNre  academic  year  
  • 13.
    Examine  Data  for  Outliers  
  • 14.
  • 15.
  • 17.
    Average  Usage  Change  By  Discovery  &  Publisher  
  • 18.
    Analyzing  Usage  Change:  %  vs  Total     Use  12   months   before   Use  12   months   aRer   %  Change   Total   Change   Journal  A   500   600   20%   100   Journal  B   5   15   200%   10   Which  is  the  beier  measure?     Is  it  the  same  for  publisher-­‐  &  journal-­‐level  data?      
  • 19.
    Reducing  varia5on  due  to  ins5tu5on  size   Currently  converNng  to  change  per  FTE       Values  are  shown  as  x  1,000  to  bring  the  change   metric  back  per  journal-­‐library  combinaNon  to  a   minimum  of  0.1     2013  JISC  Discovery  study  took  a  similar  approach  
  • 20.
    Average  Usage  Change  By  Discovery  &  Publisher   Per  Journal  &   Per  10,000  FTE  
  • 21.
    Full  Model   Including  Discovery  Service,  Publisher,  and  Library   Including  Discovery  Service,  Publisher,  and  Library  
  • 22.
    Does the effectof discovery service differ across libraries? Library  10-­‐15   Library  16-­‐21   Library  22-­‐27   Library  28-­‐33  Library  1-­‐9  
  • 23.
    Nested  ANOVA  Model     [all  three  factors  –  preliminary  results]  
  • 24.
    Does  usage  change  vary  across  libraries?     Institution (sorted by Mean Change)
  • 25.
    Does  usage  change  vary  across  publishers?     Publisher (sorted by Mean Change)
  • 26.
    Does  usage  change  vary  across  discovery  services?  
  • 27.
    Publisher Does  the  effect  of  discovery  service  differ  across  publishers?    
  • 28.
    Results   Can  we  detect  differences  between  Discovery   Services,  Publishers,  and/or  Libraries  and/or   their  interac4ons?     •  Library  –  Yes   •  Publisher  –  No   •  Discovery  Service  –  Yes   •  DifferenNal  discovery  service  effect  by   publisher  –  Yes  
  • 29.
    Next  Steps   • Design  &  test  for  effects  of:     –  Aggregator  full  text  availability   –  Publisher  Size   –  Journal  Subject   –  Overall  usage  trends  (Requires  Disc  Srvc  ‘control’)   –  ConfiguraNon  opNons  in  Discovery  services   •  Expand  pool  of  libraries   •  Perhaps  explore  WHY        
  • 30.
    Sharing  Data   • With  par5cipa5ng  libraries   –  Customized  reports  for  each  library   •  With  par5cipa5ng  publishers   –  Customized  reports  for  each  publisher   –  PresentaNons  as  requested   •  With  discovery  vendors   –  PresentaNons  as  requested   •  In  publica5ons  and  presenta5ons   –  Maintaining  anonymity  of  data      
  • 31.
    Doing  “Resarch”,  SCELC  Style!   •  Why  SCELC?   •  SCELC  Funding   –  staNsNcs  consultant     –  research  &  wriNng  retreats   –  See  hip://bit.ly/1dNMDL3  for  more  detail   •  SCELC  libraries  encouraged  to  parNcipate  in  next  round   –  Survey:  hip://bit.ly/DSparNcipaNon