Measure for Measure: The role of metrics in assessing research performance - Society for Scholarly Publishing. June 2013
1. Measure for Measure
The role of metrics in assessing research performance
Society for Scholarly Publishing - June 2013
MichaelHabib, MSLS
Product Manager, Scopus
habib@elsevier.com
Twitter: @habib
http://orcid.org/0000-0002-8860-7565
2. 1. What level am I assessing?
– Article, Journal, Researcher, Institution, etc.
Which metric should I use?
3. 1. What level am I assessing?
– Article, Journal, Researcher, Institution, etc.
2. What type of impact am I assessing?
– Research, Clinical, Social, Educational, etc.
Which metric should I use?
4. 1. What level am I assessing?
– Article, Journal, Researcher, Institution, etc.
2. What type of impact am I assessing?
– Research, Clinical, Social, Educational, etc.
3. What methods are available based on above?
– Metrics: Citation, Usage, Media, h-index, SNIP, SJR, etc.
– Qualitative: Peer-Review, etc.
Which metric should I use?
8. Background & approach
Who & when: 54,442 individuals were randomly selected from Scopus.
They were approached to complete the study in October2012.To ensure
an unbiasedresponseElsevier’sname was only revealed atthe end of
the survey.
Responses:The online survey took around 15-20 minutes to complete.
3,090 respondentscompletedit, representing a response rate of 5.7%.
Data has not beenweighted. There was a representative response by
country and discipline.
Statisticaltesting: Error margin ± 1.5%, at 90% confidencelevels.When
comparing the score formain group and sub-groups we have used a Z test
of proportion to identify differencesbetweenthe overallaverage and the
sub-group (90% confidence levels),when there are 30 or more responses.
8
Adrian Mulligan, Gemma Deakin and Rebekah Dutton
Elsevier Research & Academic Relations
9. 9
Most widely known by researchers
Impact Factor (n=2,520)* 82%
H-Index (n=1,335) 43%
Journal Usage Factor (n=309) 10%
Altmetrics (n=41)* 1%
Impact Factor is published by Thomson Reuters, Altmetrics were least well known
10. Awarenes
s
10
Q2 Which of these do you think are most useful at measuring research quality? (Select up to 3)
64%
29%
29%
28%
58%
37%
34%
42%
0% 20% 40% 60% 80% 100%
Impact factor (n=2,530)
SNIP (n=51)
SJR (n=126)
Eigenfactor (n=285)
h-index (n=1,335)
Journal Usage Factor (n=309)
F1000 (n=155)
Alt-metrics (n=41)
* Only people who said they were aware of a particular metric in Q1 were given the opportunity to
select that metric in Q2, *See appendix for background and approach.Research by Elsevier Research
& Academic Relations. ImpactFactor is published by Thomson Reuters,
TOTAL
(n=3,090)
82%
2%
4%
9%
43%
10%
5%
1%
Researcher perception of most useful
% useful
11. Generally, metrics with the highest awareness are
also considered to be the most useful
11
Impact factor
SNIP
SJR
Eigenfactor
h-index
Journal Usage Factor
F1000
Alt-metrics
R² = 0.6972
0%
10%
20%
30%
40%
50%
60%
70%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%
Percentageofawarerespondentsthatchosethe
metricasoneofthemostuseful
Percentage of respondents that are aware of the metric
The trendline showsthe linear
trendfor the relationship
betweenawarenessandusage
of metrics
Metricsabove the line have
lowerlevelsof awareness,but
are more likelytobe ratedas
useful thanthe typical
awareness-usage relationship
Metricsbelowthe line have
higherlevelsof awareness,
butare lesslikelytobe rated
as useful thanthe typical
awareness-usage relationship
*See appendix for background and approach.Research by Elsevier Research & Academic Relations.
ImpactFactor is published by Thomson Reuters,
12. 13
Assessing the usefulness of potential quality
metrics: by age
Significantdifferencebetween
subsetandtotal (subset higher)
Significantdifferencebetween
subsetandtotal (subset lower)
Under 36 (n=540) 36-45 (n=920) 46-55 (n=819) 56-65 (n=507) Over 65 (n=242)
TOTAL
(n=3,090)
Article
views/downloads (for
articles)
43%
Citations from
materials thatarein
repositories
43%
Share in social
network mentions (for
articles)
16%
Number of readers
(for articles) 40%
Number of followers
(for researchers) 31%
Votes or ratings (for
articles) 24%
A metric that measures
the contribution an
individual makes to peer
review (for researchers)
28%
A score basedon
reviewer assessment (for
articles)
28%
Q3 Thinking about possible new measures of research productivity, how useful do you think the below would be in assessing the quality of a
researcher or a researcharticle?(By age) % Think it would be extremely/veryuseful
43%
49%
21%
42%
38%
35%
34%
33%
44%
45%
18%
41%
33%
24%
29%
29%
45%
41%
15%
39%
28%
22%
27%
27%
44%
41%
12%
41%
30%
22%
26%
27%
36%
37%
13%
35%
30%
19%
24%
27%
13. 14
Assessing the usefulness of potential
quality metrics: by region (1 of 2)
Significantdifferencebetween
subsetandtotal (subset higher)
Significantdifferencebetween
subsetandtotal (subset lower)
Africa
(n=72)
APAC
(n=803)
Eastern Europe
(n=183)
Latin America
(n=182)
TOTAL
(n=3,090)
Articleviews/downloads (for
articles) 43%
Citations frommaterials thatare
in repositories 43%
Share in social network mentions
(for articles) 16%
Number of readers (for articles) 40%
Number of followers (for
researchers) 31%
Votes or ratings (for articles) 24%
A metric that measures the
contribution an individual makes
to peer review (for researchers)
28%
A scorebased on reviewer
assessment(for articles) 28%
Q3 Thinking about possible new measures of research productivity, how useful do you think the below would be in assessing the quality
of a researcher or a research article? (By region, slide 1 of 2) % Think it would be extremely/very
useful
56%
51%
26%
49%
36%
33%
40%
44%
50%
55%
27%
46%
46%
29%
35%
36%
50%
49%
19%
45%
41%
30%
28%
26%
50%
49%
21%
45%
34%
24%
32%
35%
14. 15
Assessing the usefulness of potential quality
metrics: by region (2 of 2)
Significantdifferencebetween
subsetandtotal (subset higher)
Significantdifferencebetween
subsetandtotal (subset lower)
Middle East (n=47) North America (n=770) Western Europe (n=1,033)
TOTAL
(n=3,090)
Articleviews/downloads (for
articles) 43%
Citations frommaterials that
are in repositories 43%
Share in social network
mentions (for articles) 16%
Number of readers (for articles) 40%
Number of followers (for
researchers) 31%
Votes or ratings (for articles) 24%
A metric that measures the
contribution an individual
makes to peer review (for
researchers)
28%
A scorebased on reviewer
assessment(for articles) 28%
Q3 Thinking about possible new measures of research productivity, how useful do you think the below would be in assessing the quality of
a researcher or a research article? (By region, slide 2 of 2) % Think it would be extremely/very
useful
40%
40%
19%
43%
32%
28%
32%
34%
41%
42%
10%
36%
23%
19%
26%
26%
36%
32%
11%
36%
23%
22%
23%
22%
15. “For publishers
Greatly reduce emphasis on the journal
impact factor as a promotional tool, ideally by ceasing to
promote the impact factor or by presenting the
metric in the context of a variety of journal-
based metrics … that provide a richer view of journal
performance.”
– from The San Francisco Declaration on Research
Assessment(DORA)(http://am.ascb.org/dora/ )
16. Transparency: Calculated by independent third-parties; freely and publicly
accessible www.journalmetrics.com
Subject Field Normalization: allows for comparison independent of the
journals’ subject classification. Reflects most current journal scopes, thereby taking
ongoing changes into account
3-year citation window: demonstrably the fairest compromise
Manipulation-resistant: Article type consistency. Only citations to and from
articles, reviews, and conference papers are considered
Breadth of coverage: Scopus has over 20,500 sources: 19,500 journals as well
as trade publications, proceedings and book series.
Metrics based on Scopus.com: underlying database available for
transparency; Titles indexed based on transparent criteria by independent advisory
board
CONS:
More complex methodology
Do not take amount of review content into account
Low awareness
Advantages of SNIP & SJR
17. Modified SNIP
• Refined metric calculation,bettercorrects
for field differences
• Outlier scores are closer to average
• Readilyunderstandablescoringscale with
an average of 1 for easy comparison
Modified SJR
• More prestigious nature of citationsthat
come from within the same, or a closely
related field
• Overcome the tendency for prestige
scores the quantityof journalsincreases
• Readily understandablescoring scale
with an average of 1 for easy
comparison
http://www.journalmetrics.com/
18. + + +
A journal’s raw
impact per paper
Citation potential in
its subject field
Peer reviewed
papers only
A field’s frequency
and immediacy
of citation
Database
coverage
Journal’s scope
and focus
Measured relative to
database median
SNIP: Source-normalized impact per paper
20. APIs are available to easily + freely embed these
metrics on your journal homepages
21.
22. Snowball Metrics …
• Support universities’ strategic decision
making processes
• Aim to encompass the entire scope of key
research and enterprise activities
• What is the driver? – universities’ metrics tend to
suit their data and priorities. With Snowball, they
agree to a single method so that they can
benchmark themselves against their peers
• What is special about them?
– owned by distinguished universities, including Oxford
and Cambridge, and not imposed by e.g. funders.
Universitiestaking control of their own destiny!
– Triedand tested methodologiesthat are available
free-of-charge to the higher education sector
– Academia– industry collaboration
23
23. Vision for Snowball Metrics
Snowball Metrics drive quality and efficiency across higher
education’s research and enterprise activities, regardless of
system and supplier, since they are the preferred standards used
by research-intensive universities to view their own performance
within a global context
24
Snowball Metrics Project Partners
24. Snowball Metrics Recipe Book
25
Agreed and tested methodologies for new Snowball
Metrics, and versions of existing Snowball Metrics, are
and will continue to be shared free-of-charge.
None of the project partners will at any stage
apply any charges for the methodologies.
Any organisation can use these methodologies for
their own purposes, public service or commercial.
(Extracts from Statement of intent,October2012)
www.snowballmetrics.com/metrics
25. 1. Choose methods + metrics appropriate to level and impact type
being assessed (DORA)
2. Don’t confuse level with type (alms ≠ altmetrics)
Free + easy to embed Scopus Cited-by counts on article pages
http://www.developers.elsevier.com/
3. Awareness of metrics correlates to acceptance, raising awareness
matters
4. APAC + younger researchers open to new metrics
5. Don’t use just one metric, promote a variety of metrics
Free + easy to embed SNIP/SJR on journal homepages
http://www.journalmetrics.com/
6. Choose transparent and standard methods + metrics
Learn more about Snowball Metrics
http://www.snowballmetrics.com/
In summary