Review Article

 

Does the READ Scale Work for Chat? A Review of the Literature

 

Adrienne Warner

Assistant Professor, Learning Services Librarian

University of New Mexico

Albuquerque, New Mexico, United States of America

Email: adriennew@unm.edu

 

David A. Hurley

Discovery and Web Librarian

University of New Mexico

Albuquerque, New Mexico, United States of America

Email: dah@unm.edu

 

Received: 29 Mar. 2021                                                                  Accepted: 13 May 2021 

 

 

cc-ca_logo_xl 2021 Warner and Hurley. This is an Open Access article distributed under the terms of the Creative CommonsAttributionNoncommercialShare Alike License 4.0 International (http://creativecommons.org/licenses/by-nc-sa/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly attributed, not used for commercial purposes, and, if transformed, the resulting work is redistributed under the same or similar license to this one.

 

DOI: 10.18438/eblip29947

 

 

Abstract

 

Objective – This review aims to determine the suitability of the READ Scale for chat service assessment. We investigated how librarians rate chats and their interpretations of the results, and compared these findings to the original purpose of the Scale.

 

Methods – We performed a systematic search of databases in order to retrieve sources, applied inclusion and exclusion criteria, and read the remaining articles. We synthesized common themes that emerged into a discussion of the use of the READ Scale to assess chat service. Additionally, we compiled READ Scale designations across institutions to allow side-by-side comparisons of ratings of chat interactions.

 

Results – This review revealed that librarians used a variety of approaches in applying and understanding READ Scale ratings. Determination of staffing levels was often the primary goal. Further, librarians consistently rated chat interactions in the lower two-thirds of the scale, which has implications for service perception and recommendations.

 

Conclusion – The findings of this review indicated that librarians frequently use READ Scale data to make staffing recommendations, both in terms of numbers of staff providing chat service and level of experience to adequately meet service demand. Evidence suggested, however, that characteristics of the scale itself may lead to a distorted understanding of chat service, skewing designations to the lower end of the scale, and undervaluing the service.

 

 

Introduction

 

Researchers have been strategizing for decades about how best to capture what happens in reference interactions. The Reference Effort Assessment Data (READ) Scale (Gerlich & Berard, 2007) was one of several tools developed in response to librarians’ “deep dissatisfaction” (Novotny, 2002, p. 10) with the reference statistics being collected in the early 2000s. The then-common practice was to record simple counts, often as hash marks on paper, for each mode (e.g., desk or telephone) in one of two categories: “directional” or “reference”. The READ Scale, by contrast, is a six-point scale indicating the amount of effort a librarian expends on each reference interaction. Answers are rated 1 if they require no specialized knowledge or consultation of resources. A rating of 6 indicates “staff may be providing in-depth research and services for specific needs of the clients” (Gerlich, n.d.). In introducing the READ Scale, Gerlich and Berard (2007) emphasized that their goal was to change the focus from “how many” and “what kind” to the knowledge, skills, and abilities needed to provide the service. They suggested this data could be used for “a retooling of staffing strategies,” to “increase positive self-awareness of the professional librarian” (2007, p. 9) and “for training and continuing education, renewed personal and professional interest, and reports to administration" (Gerlich & Whatley, 2009, p. 30). While other classification systems exist (see Maloney & Kemp, 2015, for examples), the READ Scale has become a standard in many libraries—a state that is reflected in its integration into both open source and commercial reference management products (e.g., Sarah, 2012).

 

Gerlich and Berard (2010) undertook a large-scale, multi-institution viability study in 2007. However, the data reflected the predominantly face-to-face nature of reference at that time. With 15 institutions reporting 8,439 transactions, 91% were face-to-face. Chat transactions accounted for only 1% of the interactions in the study. The landscape has shifted dramatically since then. Chat is now a major source of reference transactions in academic libraries (Asher, 2014; Belanger et al., 2016; Nicol & Crook, 2013; Ward & Phetteplace, 2012), accounting for more than 20 percent of all reference transactions at the University of New Mexico in 2019, with in-person reference transactions down to just 55 percent. In 2020, the majority of reference transactions were conducted via chat, likely due to COVID-19 related building closures.

 

The software that mediates chat reference interactions typically generates rich metadata, from timestamps to transcripts, allowing, and likely encouraging, innumerable assessment strategies. Indeed, chat reference evaluation has been the focus of hundreds of research articles since its first implementation in academic libraries in North America in the mid 1990s (Matteson et al., 2011), with metrics including number of chat interactions in a given timeframe, number of missed chats, frequency and length of interaction, turns taken between librarian and user, word counts, and type of referring URL (Luo, 2008). These automatically generated metrics are often combined with more qualitative measures such as types of questions asked; the presence of reference interviews and instructional elements; and quality, completeness, and tone of librarians’ answers (Luo, 2008).

 

Some chat-specific, or at least virtual reference specific, assessment tools have been developed, both for services overall (Hirko, 2006; White, 2001), and for transcript analysis in particular (Mungin, 2017). However, transcript analysis is both time consuming (Mungin, 2017) and limited (Belanger et al., 2016; Rabinowitz, 2021). Much of the literature on chat reference has been case studies with limited generalizability, with McLaughlin (2011) suggesting the need for standard approaches and reporting formats across libraries.

 

There has been recognition generally that standards and assessments are not necessarily transferable from face-to-face to online reference (Ronan et al., 2003), and some researchers have attempted to determine applicability across modes (Schwartz & Trott, 2014). However, though the READ Scale is widely used with chat reference, and is recommended as a metric "applicable across reference services" in the Reference and User Services Association's Guidelines for Implementing and Maintaining Virtual Reference Services (2017), no equivalent to the viability study has been undertaken for chat. Library services have changed considerably in the nearly two decades since the READ Scale was developed, and it cannot be assumed that a tool developed for in-person contexts will then be appropriate for chat reference in 2021. In that sense, there is no evidence that the READ Scale is appropriate for chat. As a first step in investigating the viability of the READ Scale for chat, we review the literature on chat reference that uses the READ Scale.

 

Aims

 

There are several ways in which a literature review can help us determine the suitability of the READ Scale for chat. First, we gain insight into how and why librarians are using the READ Scale; that is, we want to see what librarians using the READ Scale for chat reference are trying to understand and what decisions they are trying to make with READ Scale data. We are interested if these are the same uses that Gerlich and Berard (2007) anticipated, and if the nature of READ Scale data is appropriate for these uses. Second, we can examine the ratings that are reported for chat reference across institutions, much as Gerlich and Berard did for all reference transactions. Here we are interested in the ratings themselves, including what patterns are evident, but also in any modifications being done to the READ Scale, as well as how the data are interpreted and reported in the literature. We use “librarians” to refer to all library workers who provide or analyze chat service. We use “chat agents” to describe the workers who field chat in situ, distinguished from those who later analyze trends.

 

Thus, we have two broad questions to guide our review:

 

        How do librarians use the READ Scale to assess chat?

        How do librarians rate chats on the READ Scale?

 

Taken together, these two strands present a picture of the READ Scale in practice, and whether or not it is an appropriate tool for the assessment of chat reference.

 

Method

 

We took a systematic approach to collecting pertinent professional literature using a mixed methods review synthesis (Heyveart et al., 2016). First, we explicitly defined inclusion and exclusion criteria for the literature we would review. Then we developed a search strategy to locate literature that would meet our inclusion criteria. Next, we read the literature to identify themes and patterns, and iteratively reread and assigned categories until we reached consensus on both the categories that were present in the literature, and the specific categories present in each document.

 

Inclusion and Exclusion Criteria

 

We defined “professional literature” to include articles, white papers, book chapters, and conference proceedings and presentations. For clarity, we use the term “article” to refer to all document types. To be included, articles must have directly discussed the READ Scale as applied to chat reference, other than reporting data from a different source, such as in the literature review section of an article. Exclusion criteria were review articles that did not present otherwise unpublished data and articles in which READ Scale data and discussion of chat cannot be distinguished from other modes of reference. We aimed for a global scope and so explicitly did not exclude material based on publication language or library type.

               

Search Strategy

 

We searched Library & Information Sciences Abstracts (LISA), Library, Information Science & Technology Abstracts (LISTA), ProQuest Dissertations & Theses Global Full Text Global, Web of Science Core Collection, Google Scholar, EBSCO Discovery Service (EDS), and the e-LIS repository. Though platforms vary in the exact construction of searches, our search had two concepts: the READ Scale and chat reference. While we could be confident that the presence of the phrase “Reference Effort Assessment Data” referred to the scale developed for library reference services, not all relevant documents spelled out the acronym. Further complicating matters, the phrase “READ Scale” was used in ways that are not related to the READ Scale. Therefore, our search required either the use of the full name of the scale, or the acronym and a variation on the word library and the word reference. The second concept was chat reference, which we defined to include any synchronous online reference service, such as chat or instant messaging. Because e-LIS, which does not index full-text, had only one result for “READ Scale” and none for “Reference Effort Assessment Data”, we did not include that concept in the query. Instead, our query looked only for the concept of chat reference, and we manually checked each “Reference work” subject result for the READ Scale concept.

 

Table 1

Database Search Strategies

Database

Database note

Search

EDSa, LISA, LISTA, ProQuest Dissertations and Theses Full Text Global, Web of Science Core Collection

(chat* OR trigger* OR "instant Messag*") AND (("READ Scale" AND librar*) OR “Reference Effort Assessment Data”)

Google Scholar
Google allows neither truncation nor parentheses.

"Reference Effort Assessment Data" OR "READ Scale" AND library OR libraries OR librarians AND Reference AND chat OR "Instant messaging" OR "instant message"

e-LIS
e-LIS’s search interface consists of a series of fields and search operators. The Keywords field allows text entry while Subjects is a drop down menu.

Field: Keywords Any of: Chat messaging messenger
Field: Subjects Any of: ....IJ. Reference work.

a Every EDS configuration is unique. While the total number of databases at UNM is in the hundreds, the databases within our EDS that produced hits on our query were: Academic Search Complete, Applied Science & Technology Source, British Library Document Supply Centre Inside Serials & Conference Proceedings, Business Source Complete, Complementary Index, Directory of Open Access Journals, Education Research Complete, ERIC, eScholarship, Gale Academic OneFile, Gale OneFile: Computer Science, Library, Information Science & Technology Abstracts, ScienceDirect, Social Sciences Citation Index, and Supplemental Index.

 

Themes and Patterns

 

We read each article and noted patterns and themes related to our guiding questions, including descriptions of how READ was used, the READ score data itself, and any resultant service outcomes. We were interested in similarities across institutions as well as differences in implementation or interpretation.

 

Results

 

Our search strategy yielded 141 unique items. After applying the inclusion and exclusion criteria to the search results, we had a total of 18 articles that we included in the review. All were from academic institutions, of which two were outside the United States. The data, patterns, and themes found in those articles are presented below.

               

Rating Comparability

 

Gerlich and Berard (2010), as part of testing its viability across many institutions, normed   READ Scale designations with coordinators at each institution who then normed the Scale with their local reference agents. Larson et al. (2014) highlighted consistent application of the scale as an issue when a chat service employed both local and consortial chat agents. However, the rest of the articles did not have any cross-institution norming, though five (in addition to Gerlich and Berard), performed some form of inter-rater norming for their own data (Belanger et al., 2016; Kemp et al., 2015; Keyes & Dworak, 2017; Maloney & Kemp, 2015; Stieve & Wallace, 2018). In the remainder, no mention was made of any norming of READ Scale ratings, other than Warner et al. (2020) who explicitly stated that interrater reliability was not tested as part of the study.

 

Rating Process

 

Also potentially limiting the comparability across institutions is who rated the chats. In eight articles, the ratings were applied post-hoc by the researchers, in three cases, the reference agents rated their own interactions, while in six articles, it is not stated who rated the chats. Kohler (2017) rated chats algorithmically, and compared the ratings to those done by the chat agents, concluding that the algorithm was as good or better at rating the effort required by the chat agents than the agents themselves.

 

Rating Questions or Answers

 

Asher (2014), Cabaniss (2015), Maloney and Kemp (2015), Ward and Phetteplace (2012), and Ward and Jacoby (2018) used the READ Scale to rate incoming questions, while others looked at the outbound responses of the librarians: the University of Turin used the ratings to “uniformly categorize the type of responses our patrons received” (Bungaro et al., 2017, p. 4). Kohler (2017) reported Rockhurst University used the READ Scale to understand “the effort, skills, knowledge, teaching, techniques, and tools” used by librarians (p. 138). Keyes and Dworak (2017), Mavodza (2019), Stieve and Wallace (2018), Valentine and Moss (2017), and Warner et al. (2019) did not address whether they were assessing the complexity of either questions or answers.

 

Local Adaptations

 

Five institutions represented in the literature adjusted the READ Scale. Kohler (2017) added a 0 rating to algorithmically rated transactions that the algorithm was not able to rate. While these were largely staff demonstrations, they were included in the analysis. Belanger et al. (2016) dropped level 6 of the scale before assessment started, though no rationale was given. Keyes and Dworak (2017) shifted the scale to 0-5, without explanation. No articles mentioned adjustments to extend the scale beyond level 6.

 

Kayongo and Van Jacob (2011) added 26 sub-categories to the READ Scale, e.g. within level 3, there are categories such as “L3 Complex known item search [Do we own?],” “L3 Online resource problem,” and “L3 Simple citation verification”. Stieve and Wallace (2018) added the word “circulation” as a bullet point in the definition of level 2 and expanded several of the examples that accompany the definitions.

 

Local Interpretations

 

Specific READ Scale ratings were not always interpreted to mean the same level of expertise. Belanger et al. (2016) defined points 3 and above as “requiring a complex response” (p. 12). Bungaro et al. (2017) only reported the split between the highest 3 categories (i.e., 4, 5, and 6), for which a subject specialist was justified and the lowest 3, for which a 'generic' librarian was sufficient. Kemp et al. (2015), comparing complexity of questions across modes, considered questions “complex” at ratings 3 and above and “basic” at 2 and below. They made a further distinction that a score of 4 or above required librarian or subject librarian expertise. Using a 0-5 scale, Keyes and Dworak (2017) noted that 0-2 were considered to be “clearly” able to be addressed by graduate students. In presenting the data, they also grouped 4 and 5, but without explanation. Maloney and Kemp (2015) grouped 4 and above as requiring advanced expertise.

 

While Mavodza (2019) reported counts for each point on the scale, she presented the groupings of the counts at 2 and lower, writing “[w]hen analyzing the same chats from the READ Level, most of them were at the one and two difficulty levels at 939 and 230, respectively” (p. 128).  Even though level 3 had only 40 occurrences fewer than level 2, she grouped the level 3 count (n=190) with level 4 (n=98). She grouped the level 5 (n=26) and 6 (n=15) together. These groupings suggest both affinities of content within the groups as well as an intellectual divide between the groups.

 

Data Points Used in Conjunction with READ Scale

 

Many times, the READ Scale ratings were used in conjunction with other data points, time measures being most common. Belanger et al. (2016), Gerlich and Berard (2010), and Maloney and Kemp (2015) each analyzed the READ Scale ratings by point in the academic semester or quarter in order to understand the relationship between the academic calendar and the complexity of reference transactions. Cabaniss (2015), Kayongo and Van Jacob (2011), and Ward and Jacoby (2018) compared ratings across times of day in order to understand busyness patterns by hour. Cabaniss (2015) also looked at days of the week in order to find the busiest and least busy days.

               

The second most-frequent variable used with the READ Scale was the topic of the interaction. Bungaro et al. (2017) were interested in the frequency of psychology-related chats and whether they were, on average, more complex than others. Cabaniss (2015) and Mavodza (2019), following the same research protocol, assigned four categories to chats: general information, technical, known item lookup, and reference. Kohler’s (2017) algorithm weighed certain words more than others in order to algorithmically assign a READ Scale rating. Ward and Jacoby (2018), studying referrals given in chat, compared the READ Scale to their own categories of referral needed, referral provided, appropriate referral, and referral gap. They found that referrals happened more often as the complexity of the question increased, and also found that the referral gap, or rate at which a referral was warranted but not given, also went up with the complexity of the interaction as designated by the READ Scale.

 

Other metrics used in conjunction with the READ Scale to determine service patterns included staffing types (Keyes & Dworak, 2017), delivery mode (Asher, 2014; Gerlich & Berard, 2010; Maloney & Kemp, 2015; Ward & Jacoby, 2018), length of interaction, and referral type (Ward & Jacoby 2018).

 

Goals of READ Scale Assessment

 

In some cases, such as in Larson et al. (2014), the READ Scale chat data were reported as part of an overall assessment of reference services, for which there was no particular insight or assessment-related decision that the authors were using the READ Scale to understand. In one case, the READ Scale ratings were used as a cut off for including transactions in a different assessment effort (Valentine & Moss, 2017). Gerlich and Berard (2010) and Belanger et al. (2016) were interested in assessing the READ Scale itself, the latter concluding that the READ Scale did not provide a sufficiently nuanced understanding of the complexities of chat service provision.

 

Several uses for the READ Scale were found in multiple articles, and these are summarized below.

 

Staffing Needs

 

Eleven articles used the READ Scale to determine appropriate staff allocation, with the most common approach being to match READ Scale ratings with the level of expertise required of the reference staff. Often the six-point scale was reduced, in the data analysis, to two broad categories: ratings that indicate a need for professional expertise, and those that do not. This sort of grouping is implied by the logo for the READ Scale, which depicts lines separating 5 and 6 from the lower numbers on the scale. This division is not explicit in the scale’s definition, however, and as discussed above, institutions located this split at varying points on the scale.

 

Researchers found evidence to support a variety of staffing recommendations, both in terms of number of staff assigned to providing the service and the level of experience of staff providing the chat service. Kemp et al. (2015) and Maloney and Kemp (2015), studying the implementation of a proactive chat system at the University of Texas-San Antonio library, found an increase in complex questions through the proactive system. Because they defined READ Scale 4 and above as needing librarian-level attention, an increase in librarians was needed to field the more frequent, complex questions, and they subsequently used a multi-pronged approach to get more librarians into the chat staffing mix (Kemp et al., 2015). Similarly, Bungaro et al. (2017) found evidence to maintain and keep subject librarians answering a subject-specific chat channel, whereas they assumed generalist librarians could adequately answer READ Scale 1-3. Further evidence was found to keep professional librarians answering chat, arguing that outsourcing late night chat service to consortial librarians resulted in decreased quality of service (Kayongo & Van Jacob, 2011).

 

While some researchers found evidence for staffing increases, Ward and Jacoby (2012) found that a rearrangement of staff to match hour-by-hour patterns in chat may optimize which staff are most likely to receive the most complex questions. However, the researchers ultimately recommended a reduction in the experience level of chat staff, shifting to graduate students exclusively. In another study, Keyes and Dworak (2017) found evidence to recommend a reduction in professional level/experience. Keyes and Dworak noted that undergraduate students could adequately field chat at the READ Scale 1-3 level.

 

While many supported point-of-need staffing with experienced or professional librarians, some suggest the tiered model was adequate to address complex questions that could not be answered at the time they were asked. In their study examining the rates of referrals and whether they occurred as needed, Ward and Phetteplace (2012) found that while the rates of referral went up with the READ Scale designation, so did the rate of interactions that should have included a referral but did not. The referral ecosystem was further complicated with the READ Scale when staff members sometimes referred questions they rated as complex, but gave little effort to answer on the spot (Belanger et al., 2016).

 

Comparison of Reference Configurations

 

Researchers in five articles used the READ Ratings to compare different types of reference service. Four of the five compared passive and proactive chat configurations (DeMars et al., 2018; Kemp, et al., 2015; Maloney & Kemp, 2015; Warner et al., 2020). Each of these researchers found that proactive chat systems resulted in more complex transactions than passive chat configurations. Stieve and Wallace (2018) compared sources of chat transactions, finding that READ Scale ratings were higher from within the university’s learning management system as compared to chats originating from the library website.

 

Gerlich and Berard (2010) compared all modes of reference at the participating institutions: walk-up directional, walk-up reference, phone directional, phone reference, email, and chat. The difference between “directional” and “reference” phone and walk-up modes was not explained. Chat transactions made up only about 1% of the transactions in the study. A greater emphasis was on distinguishing between “walk up” in-person desk interactions and those in-person interactions that happened in hallways and offices: “The off-desk comparisons show… that the percentage of questions answered off-desk for most of the institutions require a much higher level of effort, knowledge, and skills from reference personnel than at the public service point” (Gerlich & Berard, 2010, p. 125)

 

Other

 

Several researchers noted the higher degree of difficulty in fielding chat rather than face-to-face interactions, difficulties that were not represented in the READ Scale. Chat agents may have to field multiple chats at the same time (Cabaniss, 2015; DeMars, 2018; Keyes & Dworak, 2017), and the necessity of typing succinct directions without the ability to rely on non-verbal cues was challenging (Gerlich & Berard, 2010). Librarians at one institution reported that multiple factors coalesced into chat interactions being deemed as more stressful (Ward & Phetteplace, 2012).

 

The Ratings

 

While the READ Scale is a 6-point scale, the 1-4 range of the scale was heavily used, while the 5-6 range was not. Belanger et al. (2016), Gerlich and Berard, (2010), Kohler (2017), Stieve and Wallace (2018), and Ward and Jacoby (2018) found that zero chats were rated a 5 or 6. Kayongo and Van Jacob (2011), Cabaniss (2015), and Mavodza (2019) found that the majority of chat interactions fell within the 1-3 range. The one outlier was Bungaro et al. (2017), who found that almost half (46%) of their chat interactions happened at the 4 or above designation, though it is unclear how many were rated at each level.

 

Not every source provided a breakdown by each number, with some providing percentages of ranges of READ Scale designations. However, of those that did, the data showed the vast majority of transactions occurring in the 1-4 range. Within the READ Scale 1-4 range, most chats are rated a READ Scale 2 or 3 (see Figure 1).

 

Table 2

Summary of Articles, Including READ Scale Breakdowns

Author, Publication Date, (Institution examineda)

Chat sample size

Breakdown by READ Scale designation as reported by sourceb

Timeframe of chats assessed with READ Scale

Sampling Method

Asher, 2014

(University of Indiana, Bloomington)

149

unstated

2006-2013

Convenience

Belanger et al., 2016

(University of Washington)

3721 chat transcripts

1=10%

2&3= 80%

4=10%

5=0

6=n/a

 

Fall quarter 2014 (September-December)

Convenience

Bungaro et al., 2017

(University of Turin; Italy)

121

1-3= 53.8% (n=65)

4-6= 46% (n=56)

2014-2016

Convenience

Cabaniss, 2015

(University of Washington)

608

1=28% (n=169)

2=45% (n=275)

3=23%(n=142)

4=4%(n=27)

5=0% (n=2)

6=unstated

3 sample weeks in winter term 2014 (January-March)

Convenience

DeMars et al., 2018

(California State University, Long Beach)

unstated

1=7-14%

2=32-32%

3=38-40%

4=7-15%

5=1-3%

6=.5-1%

Gleaned from table

2016-2018. Compared three chat configurations.

Cluster

Gerlich and Berard, 2010

(Multiple)

Test 1: 98 Test 2: 317

Test 1:

1=13% (n=13)

2=19% (n=19)

3= 45% (n=44)

4=22% (n=22)

5=0% (n=0)

6=0% (n=0)

 

Test 2:

1=6% (n=19)

2=24% (n=76)

3=47% (n=150)

4=21% (n=66)

5=2% (n=6)

6=0% (n=0)

Test 1: 3 weeks in February, 2007

Test 2: Spring semester, 2007

Test 1: Convenience, Test 2: Adaptive

Kayongo and Van Jacob, 2011

(University of Notre Dame)

2517

1=21.1% (n=531)

2=28.3% (n=712)

3=40.4% (n=1017)

4=5.9% (n=1480

5=4.2% (n=105)

6=.1% (n=4)

November 2007-May 2010

Convenience

Kemp et al., 2015

(University of Texas, San Antonio)

 

Test 1: unspecified

Test 2: 287

Test 3: 228

Test 1:

3= 44%

4 and above= 21%

 

Test 2: (triggered)

1,2 =19%

3 and above= 81%

 

Test 3: (non-triggered)

1,2=37%

3 and above= 63%

Test 1:

Six sample weeks from Fall 2013 to Spring 2014.

Tests 2 and 3: November 2013

Test 1:

Convenience

Tests 2 and 3:

Cluster

Keyes and Dworak, 2017

(Boise State University)

454

0-1= 19% (n=78)

2= 31% (n=131)

3= 39% (n=164)

4-5= 11% (n=48)

May 2014-Sept 2016

Convenience

Kohler, 2017

(Rockhurst University)

1109

0=7% (n=80)

1=15% (n=166)

2=33% (n=366)

3=41% (n=456)

4=3% (n=35)

5=1% (n=6)

6=0% (n=0)

FY2015- FY2016

Convenience

Larson et al., 2014

(University of Maryland)

39

1=5% (n=2)

2=49% (n=19)

3=21% (n=8)

4=23% (n=9)

5=3% (n=1)

6=0% (n=0)

2013

Convenience

Maloney and Kemp, 2015

(University of Texas, San Antonio)

2492

1=3% (n=85)

2=31% (n=764)

3=39% (n=968)

4=26% (n=654)

5=1% (n=19)

6=0% (n=2)

6 sample weeks: Fall 2013 and Spring 2014

Cluster

Mavodza, 2019

(Zayed University; United Arab Emirates)

1498

1=63% (n=939)

2=15% (n=230)

3=13% (n=190)

4=7% (n=98)

5=2% (n=26)

6=1% (n=15)

Feb 2013-Feb 2018

Convenience

Stieve and Wallace, 2018

(University of Arizona)

382

1=0% (n=2)

2=34% (n=128)

3=48% (n=184)

4=9% (n=33)

5=3% (n=13)

6=0% (n=0)

Fall and Spring semesters for 2014 and 2015.

Cluster

Valentine and Moss, 2017

(University of Kansas)

30

3 and above=100% (n=30)

Fall semester 2016

Cluster

Ward and Jacoby, 2018

(University of Illinois, Urbana-Champaign)

 

1120

 

1=30 (3%)

2=378 (34%)

3=636 (57%)

4=76 (7%)

5 and 6=unstated

April 2015

Convenience

Ward and Phetteplace, 2012

(University of Illinois. Urbana-Champaign)

unspecified

Unspecified

 

Sept and Oct 2010

Convenience

Warner et al., 2020

(University of New Mexico)

4617

unspecified

July 2016 to July 2018

Cluster

a  Country of institution is United States of America unless otherwise indicated.

b When the source gives numbers only, review authors have converted these to percentages and rounded to the whole number, but we have not used percentage-only data to derive hard numbers. Because of this, percentages are recorded in the 100-101 range.

 

Figure 1

READ Scale ratings for chat interactions from articles that provided breakdowns by each READ Scale number. Gerlich and Berard (2010) reported two tests in “Testing the Viability of the READ Scale.”

 

Discussion

 

Perhaps unsurprisingly, a primary goal in using the READ Scale at many institutions was to understand staffing needs. Many analyzed READ Scale designations by time of day, day of week, and time in the academic period in order to find patterns of busiest times and most complex questions, following assessment trends Luo identified in 2008. Understanding demand for service is an admirable goal, but unfortunately even when service patterns are present, staffing models may not provide the flexibility to checkerboard staff shifts to match them. And, while optimizing staffing may be possible to a degree, in practice, complex questions can arrive at any point in the day, week, or year and service coordinators must determine what is adequate, rather than optimal, service. Additionally, READ scores were used to make inferences about the level of expertise needed, most often a recommendation for decreased expertise, and none of the sources articulated how they would determine when they had sub-adequate staffing levels on chat. This use of the READ Scale, while seemingly common, is troubling.

 

The most striking feature of the READ Scale data was the paucity of chat transactions rated 5 or 6. Indeed, this was reflected in Gerlich and Berard’s (2010) viability study data. None of the 15 testing institutions reported any chat transactions falling into the 5 or 6 ratings. However, it would be premature to conclude that patrons are not asking complex questions, based solely on ratings of the responses to those questions. As Gerlich and Berard developed the READ Scale, they theorized that more complex questions were answered away from the reference desk, such as in hallways or offices, and they built this assumption into the scale. For example, the definitions provided for the 5 and 6 ratings assume or presuppose the interaction occurring outside of initial reference contexts. Level 5 suggests that “consultation appointments might be scheduled” (Gerlich, n.d.) while level 6 specifies that “(r)equests for information cannot be answered on the spot” (Gerlich, n.d.). In practice, this makes the higher end of the scale unavailable to chat transactions. As of this writing, it is not feasible for a librarian to initiate a chat with a specific patron or schedule a consultation via chat, so any question that requires follow up is necessarily answered in a different mode. The scale gives weight to the mode in which the librarian answers reference questions, rather than the mode in which the patron asks it.

 

Depending on how the READ Scale ratings are used by an institution, the data could present a distorted view of the chat service. For example, if a question comes in via chat that is beyond the abilities of that librarian, they might create a ticket that the relevant subject specialist will claim. While the eventual answer would likely be a 5 or 6, the chat transaction would rate as a 1, as it took no effort. If the staff member used knowledge or training about who specifically to refer the question to, or instructed the patron in how to schedule a consultation, it might appropriately rate at level 2 or 3. In any case, it would be at the low end of the scale, indistinguishable from simple item searches or questions about hours. Using READ Scale data to determine the level of expertise required of chat staff is problematic when questions that staff could answer comfortably and those they lacked the expertise to answer at all are represented the same way in the data.

 

It is also notable that almost every study in this review tabulated READ Scale designations based simply on frequency, either as raw counts or percentages. This gives equal weight to all questions; imagine a one hour reference shift in which a librarian has two questions at level 4, spending 15 minutes on each working with the patrons to choose a database, develop and refine search strategies, and so on. That librarian also spends ten seconds on each answer to questions about the hours. If there are five such questions, the hour was predominantly level 1, even though less than a minute was spent on level 1 answers, and half the hour was spent on level 4 responses. This approach essentially recreates the hash mark system that the READ Scale was developed to replace, with the added misinterpretation that most questions can be answered without any specialized knowledge or skill. 

 

This unequal weight distribution is true for all modes of reference, but is especially significant for chat, where librarians report fielding multiple questions at a time. Answering several questions about hours or renewals while a patron requiring more complex assistance is running searches is efficient. An approach to assessment that devalues this practice is flawed.

 

Measuring the skills needed to answer a question makes sense when trying to assess how to staff the service. It can be less helpful when trying to assess the value of the service to our patrons. Helping a patron navigate a confusing record or misconfigured link in a database may be a simple question from the librarian’s perspective, perhaps rating a 2 on the READ Scale. From the patron’s perspective, getting access to the resource may be critically important. A low READ score might be desirable in this case: the librarian was able to solve the patron’s problem quickly and easily. Yet, the interpretation of the READ Scale may rank the worthiness of some questions or answers over others, a common trap inherent in scales (LeMire et al., 2016).

 

Given the language limiting chat to the lower end of the READ Scale, we might see a rash of adaptations of the definitions and examples, creating the ability of chat to enter into the final third of the READ Scale. However, of the few institutions that adapted the READ Scale verbiage to their local assessment environments, mode-agnostic wording was not added at the upper levels, and mode-specific language was not deleted. Because the majority of institutions did not alter the language of the Scale at all, it may be that critical analysis of the viability of the tool explicitly for chat was absent or that librarians agreed with the existing wording of the scale. If librarians adhere to the existing wording of the scale, they implicitly agree that regardless of time invested or librarian resources used, chat content will never equate to the same content delivered face-to-face in a traditional consultation. If librarians assess chat in direct comparison to face-to-face, or indeed other modes of delivery, the assessment reflects bias by creating an implicit hierarchy of service.

 

While the READ Scale language may constrict chat transactions to the lower two-thirds of the scale, experience with complex content may have a similar effect. Less experienced chat agents do not have as deep a skillset with which to call upon in chat interactions, leaving their choice of READ Scale rating constricted. So, if an institution staffs chat with less experienced undergraduate students, the READ Scale ratings should reflect that restricted ability, resulting in lower-rated chats. In contrast, highly experienced librarians have the ability to answer inquiries farther up the READ Scale. Further, the judgement of complexity is compounded by more experienced librarians’ increased familiarity and expertise: librarians who walk students through the intricacies of searching databases may be more familiar with the content and thus rank those interactions as less complex.

 

Limitations

 

The findings of this literature review are limited in several ways. Because we were attempting to understand the variety of situations in which the READ Scale has been deployed to measure chat, we did not limit the scope of material to academic journal articles. The resultant formats provide variable depth of information. For example, only half of the articles included provided the breakdown of READ Scale ratings by each of its six levels, while the others reported only clusters of ratings. Some provided in-depth explanations of READ Scale ratings results and extensive analysis, and others did not. For some of the articles, chat reference was a brief focus of a comprehensive reference service evaluation. More broadly, many institutions used the READ Scale to assess reference interactions, as evidenced by our initial searches, however it is unclear exactly how many institutions used the READ Scale to assess chat. Evidence from community colleges and additional international sources could inform future research and help to address this knowledge gap. Finally, we recognize that not all chat service coordinators publish their assessment efforts, so direct research with this population may provide insight into READ Scale reach, variations in application, and changes in the quantity and experience levels of chat agents after using it for assessment.

 

Conclusion

 

Pomerantz et al. (2008) argued that when assessing chat reference, “’Good enough’ data is better than no data when one is aware of the limitations of the data” (p. 27). The READ Scale has been adopted across the globe to assess online synchronous interactions, yet this tool was developed before chat reference became commonplace (Matteson et al., 2011). In this literature review, we set out to understand how librarians use the tool to make decisions and understand resultant data when examining chat reference service. This was a first step to understanding whether the READ Scale has withstood the test of time as chat has evolved to play a much bigger role in reference service.

 

Based on the data that have been published and reviewed herein, the READ Scale systematically undervalues chat reference transactions for many of the assessment goals for which it is used. Ultimately, the way the READ Scale is applied and interpreted with regards to chat reference is at odds with the intent of the tool. Rather than capturing the knowledge, skills, and abilities needed to provide adequate reference service, it overemphasizes the simplest, least time consuming questions, with the effect of making chat reference appear not worth the cost of staffing with experienced professionals.

 

We are not providing recommendations here about the correct level of staffing for chat reference. Our concern is that the assessment that libraries do to make those decisions are appropriate and accurate. We therefore propose a number of recommendations for practice.

 

Recommendations for Practice

 

Based on our understanding of how the READ Scale has been used by librarians to conflate complexity, experience, and value, we invite practitioners to reflect on the following recommendations:

 

1.       Use caution when comparing face-to-face, chat, email, text, and phone interactions to each other using the existing definitions and examples provided in the READ Scale documentation.

2.       Examine and update the definitions and examples of each level of the READ Scale through the lens of each reference delivery method your library provides.

3.       If you are trying to understand the expertise needed to staff the service, consider rating every question that staff could not answer as a 6. These are questions that cannot be answered “on the spot” with your current staffing model.

4.       If you are trying to understand the service overall, consider reporting the time spent on each level of answer, rather than the number or percent of questions at each level. Alternately, you could develop a system of weights for each READ Scale number in order to represent expertise needed for the length of the interaction.

 

Author Contribution Statement

 

Adrienne Warner: Conceptualization, Data Curation, Formal Analysis, Investigation, Methodology, Supervision, Visualization, Writing – original draft, Writing – review and editing David A. Hurley: Conceptualization, Data Curation, Formal Analysis, Investigation, Writing – original draft, Writing – review and editing

 

References

 

Asher, A. (2014). Who’s asking what? Modelling a large reference interaction dataset. Proceedings of the 2014 Library Assessment Conference, 52-62. https://www.libraryassessment.org/past-conferences/2014-library-assessment-conference

 

Belanger, J., Collins, K., Deutschler, A., Greer, R., Huling, N., Maxwell, C., Papadopoulou, E., Ray, L. & Roemer, R. C. (2016). University of Washington Libraries chat reference transcript assessment. https://www.lib.washington.edu/assessment/projects/chat-reference-assessment-project-2014-2016-final-report

 

Bungaro, F., Muzzupapa, M. V., & Tomatis, M. S. (2017). Extending the live chat reference service at the University of Turin - A case study. IATUL Annual Conference Proceedings, 1–12. https://docs.lib.purdue.edu/iatul/2017/challenges/1/  

 

Cabaniss, J. (2015). An assessment of the University of Washington’s chat reference services. Public Library Quarterly, 34(1), 85–96. https://doi.org/10.1080/01616846.2015.1000785

 

DeMars, M. M., Martinez, G., Aubele, J., & Gardner, G. (2018, December 7). In your face: Our experience with proactive chat reference [Conference presentation]. CARLDIG-S Fall 2018 Program, Los Angeles, CA, United States. http://eprints.rclis.org/33775/   

 

Gerlich, B. K. (n.d.). READ Scale bulleted format. The READ Scale©: Reference Effort Assessment Data. http://readscale.org/read-scale.html

 

Gerlich, B. K., & Berard, G. L. (2007). Introducing the READ Scale: Qualitative statistics for academic reference services. Georgia Library Quarterly, 43(4), 7-13. https://digitalcommons.kennesaw.edu/glq/vol43/iss4/4

 

Gerlich, B. K., & Berard, G. L. (2010). Testing the viability of the READ Scale (Reference Effort Assessment Data)©: Qualitative statistics for academic reference services. College & Research Libraries, 71(2), 116-137. https://doi.org/10.5860/0710116

 

Gerlich, B. K., & Whatley, E. (2009). Using the READ Scale for staffing strategies: The Georgia College and State University experience. Library Leadership & Management, 23(1), 26-30. https://journals.tdl.org/llm/index.php/llm/article/download/1755/1035

 

Heyvaert, M., Hannes, K., & Onghena, P. (2016). Using mixed methods research synthesis for literature reviews: The mixed methods research synthesis approach (Vol. 4). Sage Publications.

 

Hirko, B. (2006). Wally: Librarians' index to the internet for the state of Washington. Internet Reference Services Quarterly, 11(1), 67-85. https://doi.org/10.1300/J136v11n01_06

 

Kayongo, J., & Van Jacob, E. (2011). Burning the midnight oil: Librarians, students, and late-night chat reference at the University of Notre Dame. Internet Reference Services Quarterly, 16(3), 99–109. https://doi.org/10.1080/10875301.2011.597632

 

Kemp, J. H., Ellis, C. L., & Maloney, K. (2015). Standing by to help: Transforming online reference with a proactive chat system. The Journal of Academic Librarianship, 41(6), 764–770. https://doi.org/10.1016/j.acalib.2015.08.018

 

Keyes, K., & Dworak, E. (2017). Staffing chat reference with undergraduate student assistants at an academic library: A standards-based assessment. The Journal of Academic Librarianship, 43(6), 469–478. https://doi.org/10.1016/j.acalib.2017.09.001

 

Kohler, E. (2017, November 3). What do your library chats say?: How to analyze webchat transcripts for sentiment and topic extraction. Brick & Click Libraries Conference Proceedings, 138–148. https://files.eric.ed.gov/fulltext/ED578189.pdf

Larson, E., Markowitz, J., Soergel, E., Tchangalova, N., & Thomson, H. (2014). Virtual information services task force report. University of Maryland University Libraries. https://doi.org/10.13016/M2GT5FH7N

 

LeMire, S., Rutledge, L., & Brunvand, A. (2016). Taking a fresh look: Reviewing and classifying reference statistics for data-driven decision making. Reference & User Services Quarterly, 55(3), 230-238.

 

Luo, L. (2008). Chat reference evaluation: A framework of perspectives and measures. Reference Services Review, 36(1), 71–85. https://doi.org/10.1108/00907320810852041

 

Maloney, K., & Kemp, J. H. (2015). Changes in reference question complexity following the implementation of a proactive chat system: Implications for practice. College & Research Libraries, 76(7), 959–974. https://doi.org/10.5860/crl.76.7.959

 

Matteson, M. L., Salamon, J., & Brewster, L. (2011). A systematic review of research on live chat service. Reference & User Services Quarterly, 51(2), 172-190. https://doi.org/10.5860/rusq.51n2.172

 

McLaughlin, J. E. (2011). Reference transaction assessment: Survey of a multiple perspectives approach, 2001 to 2010. Reference Services Review, 39(4), 536-550. https://doi.org/10.1108/00907321111186631

 

Mavodza, J. (2019). Interpreting library chat reference service transactions. Reference Librarian, 60(2), 122–133. https://doi.org/10.1080/02763877.2019.1572571

 

Mungin, M. (2017). Stats don't tell the whole story: Using qualitative data analysis of chat reference transcripts to assess and improve services. Journal of Library & Information Services in Distance Learning, 11(1-2), 25-36. https://doi.org/10.1080/1533290X.2016.1223965

 

Nicol, E. C., & Crook, L. (2013). Now it's necessary: Virtual reference services at Washington State University, Pullman. The Journal of Academic Librarianship, 39(2), 161-168. https://doi.org/10.1016/j.acalib.2012.09.017

 

Novotny, E. (2002, September). Reference service statistics & assessment. Association of Research Libraries. https://www.arl.org/resources/spec-kit-268-reference-service-statistics-a-assessment-september-2002/

 

Pomerantz, J., Mon, L. M., & McClure, C. R. (2008). Evaluating remote reference service: A practical guide to problems and solutions. portal: Libraries and the Academy, 8(1), 15-30. https://doi.org/10.1353/pla.2008.0001

 

Rabinowitz, C. (2021). You keep using that word: Slaying the dragon of reference desk statistics. College & Research Libraries News, 82(5), 223-224. https://doi.org/10.5860/crln.82.5.223

 

Reference and User Services Association. (2017). Guidelines for implementing and maintaining virtual reference services. http://www.ala.org/rusa/sites/ala.org.rusa/files/content/resources/guidelines/GuidelinesVirtualReference_2017.pdf

 

Ronan, J., Reakes, P., & Cornwell, G. (2003). Evaluating online real-time reference in an academic library: Obstacles and recommendations. The Reference Librarian, 38(79-80), 225-240. https://doi.org/10.1300/J120v38n79_15

 

Sarah. (2012, March 21). LibAnswers- new features are live! Springshare. https://blog.springshare.com/2012/03/21/libanswers-new-features-are-live/

 

Schwartz, H. R., & Trott, B. (2014). The application of RUSA standards to the virtual reference interview. Reference and User Services Quarterly, 54(1), 8-11. https://journals.ala.org/index.php/rusq/article/view/3993/4505

 

Stieve, T., & Wallace, N. (2018). Chatting while you work. Reference Services Review, 46(4), 587–599. https://doi.org/10.1108/RSR-09-2017-0033

 

Valentine, G., & Moss, B. D. (2017, March 28). Assessing reference service quality: A chat transcript analysis [Conference presentation]. Association of College and Research Libraries Conference, Baltimore, MD.  http://hdl.handle.net/1808/25179

 

Ward, D., & Jacoby, J. (2018). A rubric and methodology for benchmarking referral goals. Reference Services Review, 46(1), 110-127. https://doi.org/10.1108/RSR-04-2017-0011

 

Ward, D., & Phetteplace, E. (2012). Staffing by design: A methodology for staffing reference. Public Services Quarterly, 8(3), 193–207. https://doi.org/10.1080/15228959.2011.621856

 

Warner, A., Hurley, D. A., Wheeler, J., & Quinn, T. (2020). Proactive chat in research databases: Inviting new and different questions. The Journal of Academic Librarianship, 46(2), 102-134. https://doi.org/10.1016/j.acalib.2020.102134

 

White, M. D. (2001). Digital reference services: Framework for analysis and evaluation. Library & Information Science Research, 23(3), 211-231. https://doi.org/10.1016/S0740-8188(01)00080-9