A “Gold‐centric” implementation of open access: Hybrid journals, the “Total cost of publication,” and policy development in the UK and beyond

This paper reports analysis of data from higher education institutions in the UK on their experience of the open‐access (OA) publishing market working within a policy environment favoring “Gold” OA (OA publishing in journals). It models the “total cost of publication”—comprising costs of journal subscriptions, OA article‐processing charges (APCs), and new administrative costs—for a sample of 24 institutions. APCs are shown to constitute 12% of the “total cost of publication,” APC administration, 1%, and subscriptions, 87% (for a sample of seven publishers). APC expenditure in institutions rose between 2012 and 2014 at the same time as rising subscription costs. There was disproportionately high take up of Gold options for Health and Life Sciences articles. APC prices paid varied widely, with a mean APC of £1,586 in 2014. “Hybrid” options (subscription journals also offering OA for individual articles on payment of an APC) were considerably more expensive than fully OA titles, but the data indicate a correlation between APC price and journal quality (as reflected in the citation rates of journals). The policy implications of these developments are explored, particularly in relation to hybrid OA and potential of offsetting subscription and APC costs.


Introduction
One important feature of the current open-access (OA) publishing environment is the coexistence of fully-OA journals and 'hybrid' subscription-OA journals. Whilst the former, such as journals published by the Public Library of Science (PLoS) or BioMed Central (BMC), produce only open-access articles, the latter, now offered by most established subscription publishers, make particular articles published in subscription journals available on an OA basis, normally on payment of a fee. Some fully-OA journals also charge a perarticle fee (commonly termed an article-processing charge or APC), whilst others may be funded through sponsorship arrangements. Whilst the majority of fully-OA journals (72% in 2014) do not charge an APC, the majority of articles published in fully-OA journals (59%) are APC funded (Crawford, 2015).
As many research funders, institutions and other stakeholders adopt policies encouraging OA, and as many authors wish to make their work openly available, they are having to decide on their position in relation to the different 'routes' to OA: OA publication in journals (also called 'Gold' OA) and deposit in OA repositories ('Green' OA) (Suber, 2012). With regard to Gold, a key policy question faced by funders in particular is their attitude to hybrid journals (Björk, 2012;Solomon & Björk, 2012). With institutions already making substantial subscription payments to publishers, APCs for hybrid journals are often seen as a second payment to the same supplier for its content and therefore perceived to be publishers 'double dipping' (Prosser, 2015;Sweeney, 2014). For this reason, a number of research funders, whilst offering to pay APCs as part of their funding, have excluded hybrid journals from such policies; examples include the European Union Gold OA pilot (OpenAIRE, 2015), the Norwegian Research Council (Frantsvåg, 2015), and the Netherlands Organization for Scientific Research (NWO, 2015).
However, in the UK, since 2012 in particular, many major research funders have introduced policies encouraging the adoption of OA with an emphasis on the Gold route, explicitly allowing payment of APCs to hybrid journals. Accompanying the introduction of these policies has been a set of funding streams available for institutions to pay APCs centrally, along with other OA costs. The funders include Research Councils UK (RCUK), representing major government-sponsored agencies, and major charitable medical research funders, such as the Wellcome Trust (Charities Open Access Fund or COAF) (RCUK, 2013;Wellcome Trust, 2014). The UK has become, therefore, an interesting test case in the impact of a Goldcentric implementation of OA including support for hybrid journals, in which it is possible to track the development of APC expenditure in relation to other costs associated with journal publications, particularly subscriptions.
In this paper we report recent work carried out to examine APC expenditure in the context of subscription costs and new administrative costs, modelling the so-called 'total cost of publication' for institutions, focusing on UK higher education institutions (HEIs). In order to carry out this modelling, we first analyse expenditure on APCs in detail, examining trends in APC expenditure over recent years and identifying the key characteristics of the APC market as it is experienced by UK HEIs. APC administration costs and subscription costs are also discussed. The analysis is then placed within an international context and its implications for future policy development discussed.

Literature review
Our previous study (Pinfield, Salter, & Bath, 2015) used the term 'total cost of publication', a term adopted from policy discussions. Specifically, the term was derived from comments by the then UK Science Minister, David Willetts (Willetts, 2014), in the UK government's response to the review of the Finch report (the report which set out recommendations which have since formed the basis of the UK's current Gold-centric approach, Finch et al., 2012).
Willetts referred to the need to "develop sustainable funding models that establish a relationship between the payment of APCs (and the costs of administering them) and subscription fees for an institution" (Willetts, 2014, p. 3). The term was used in the particular context of perceived 'double dipping' by publishers and related especially to new additional costs as experienced by institutions in a hybrid OA environment.
However, the term 'total cost of publication' used in this way is not without problems. It does not, for example, include all of the costs borne by HE institutions (for instance, existing administration costs associated with subscriptions). Nor does it take into account other costs, such as those of other stakeholders, notably publishers. Rather, it focuses specifically on additional costs experienced by institutions in managing open access. An alternative term, 'total cost of ownership', has also been used in this context (Lawson, 2015;Woodward & Henderson, 2014). However, we decided not to use that term since 'total cost of ownership' is already widely used in the context of costing particularly IT systems over their life cycle (Piscopo, Johnston, & Bellenger, 2008): using the term TCO in a publishing context would cause confusion. The term, 'total cost of publication' does not have any pre-existing uses, reducing the possibility of confusion. Understood in the limited way it appears to have been originally intended (that is, focusing on new institutional costs of APCs and administration in addition to subscriptions for published content), 'total cost of publication' is used here as a convenient label in lieu of a better term. It captures something important: the extent to which hybrid OA is impacting institutional costs in relation to the publication process.
Of the different components of the 'total cost of publication' (APCs, new administration costs, and subscriptions), most recent work has been done on APCs. Several studies have provided insights into the APC market by analysing list prices charged by publishers. Björk & Solomon (2014a), using data derived from a large sample from the Scopus database, identified the mean average APC for a number of journal types with a marked difference between them. Fully-OA journals "published by 'non-subscription' publishers" had a mean APC of $1,418; fully-OA journals "published by 'subscription' publishers" had a mean of $2,097; and hybrid journals published by 'subscription' publishers", $2,727. These APC prices are noticeably higher than those produced in other studies (Morrison, Salhab, Calvé-Genest, & Horava, 2015;Solomon & Björk, 2012), which have focussed only on fully-OA journals (based on data from the Directory of Open Access Journals, DOAJ) and excluded hybrid titles. Further work by Björk & Solomon (2015) identified a correlation between APC price and journal quality, where quality is measured by citation rates (specifically using the Source Normalised Impact per Paper, SNIP, measure). This study focused on fully-OA journals but also found indications of similar segmentation within hybrid journals (with prices also varing by discipline).
Our previous work  analysed APC prices paid by institutions from centrally-managed funds based on a sample of 23 UK higher education institutions covering the period 2007 to 2013, as part of research to calculate the 'total cost of publication'.
Centrally-managed APC payments rose sharply from 2012: this was largely attributed to changes in the policies and funding arrangements of UK research funders. Based on figures from the first quarter of 2014, we projected a continued rise of central APC payments during 2014 amounting to an expected increase of over 500% since the beginning of 2012. The mean APC paid by institutions had remained relatively stable since 2008 but there was considerable variation in APC prices paid by institutions over the period, with prices ranging from £82 to £5,280. Levels of APCs charged by single publishers also varied considerably. There was a marked difference between prices charged for APCs within fully-OA journals and those of hybrid journals (corroborating Björk & Solomon, 2014a). Well-established commercial publishers with large portfolios of subscription and hybrid journals had captured a significant proportion of the APC market, with eight of the top 10 publishers (who between them received 76% of all the APCs paid within the dataset) being from this bracket, including Elsevier (who received over 20% of the APC payments) and Wiley (15%). Our attempts to model the 'total cost of publication' (TCP) were, however, hampered by insufficiently robust data on administrative costs which meant we could not include these costs. Further work was clearly needed in this area. Nevertheless, our preliminary calculations of TCP (excluding administration costs) based on 2013 APCs and 2013-2014 subscription data for 20 HEIs showed that subscriptions were £29.4 million (90% of TCP) and APCs £3.3 million (10% of TCP) . Since then, Johnson, Pinfield, & Fosci (2015) analysed detailed administrative costs by 29 UK HEIs, and reported a figure of £88 per article to administer a paid-for Gold APC payment, although overall costs to institutions of implementing OA in line with policy requirements (including institutional policy development, communication and advocacy, and reporting) were considerably higher.
This study therefore provides a base figure for calculating the APC administrative cost component of the 'total cost of publication'.
The remaining component of the 'total cost of publication', subscriptions, is arguably the best understood. However, there have not been many studies in the published literature specifically on subscription cost data, although the nature of pricing models have been extensively discussed, particularly in relation to the purchasing bundles of electronic titles (the 'big deal') (Strieb & Blixrud, 2014). Perhaps the main reason for the paucity of empirical studies is that cost data have often been restricted because of confidentiality clauses between publishers and higher education institutions (or consortia). Interestingly, however, since our previous study , there appears to have been something of a relaxing of attitudes of libraries to sharing such data. Many UK libraries have now made their subscription information publicly available in response to freedom of information (FoI) requests (Lawson & Meghreblian, 2014b). In the US, libraries have similarly responded to FoI requests resulting in recent analyses of their 'big deal' payments (Bergstrom, Courant, Mcafee, & Williams, 2014). This greater openness should continue since many subscription deals now negotiated by consortia include clauses allowing costs to be shared for FoI enquiries (Lawson & Meghreblian, 2014a).
Using subscription and APC data gathered from UK institutions, Pinfield et al. (2015) provided a provisional analysis of the 'total cost of publication'. This informed policy discussions in the UK, including the Burgess Report (Burgess, 2015), commissioned by RCUK to review its OA policy. The measure also informed the stance developed by Jisc in the UK in relation to negotiation of multi-year deals with publishers on behalf of the academic community particularly in relation to the proposal to 'offset' APC payments against subscriptions as the basis for the deals (Jisc, 2015;Lawson, 2015). Finally, it has informed the ongoing debate in this area, particularly on the topic of 'double dipping' (Prosser, 2015;Smith, 2014a).
The 'double dipping' debate has been played out particularly intensely in the UK because of the Gold-centric nature of research funders' OA policies, which have proved controversial (Björk & Solomon, 2014a;Crotty, 2014;Prosser, 2015;Smith, 2014a;Sweeney, 2014). Whilst some from the publishing community have disputed the validity of 'double dipping', arguing that APCs and subscriptions are different services and can therefore legitimately be charged for separately (Smith, 2014b), more commonly, publishers have implicitly accepted its validity in developing 'no double-dipping' policies (Jisc Collections, 2014;Royal Society Publishing, 2013). At government and funder level, there also seems to be implicit acceptance of 'double dipping' as a legitimate concern (Hall, 2012;Sweeney, 2014;Willetts, 2014), particularly in the context of the apparent policy-based encouragement of hybrid journals. One key issue is the extent that the hybrid model can reasonably be considered to be a transitional, as proposed by Prosser (2003), and, therefore, the extent to which any additional costs associated with APCs on top of subscriptions can be considered temporary (Finch, 2014;Jubb, 2014).
Despite the controversy, UK policies appear to have encouraged uptake of OA. Recent estimates  indicate that the proportion of papers produced by UK authors which are open access is greater than global averages. For papers published in 2014, 22% of papers with at least one UK author were available in an open access form immediately, compared with the global average (19%). This was 28% after six months following publication for UK research, compared with 23% globally. After 12 months, the UK figure was 38%, and 43% after 24 months, compared with 29% and 34% over the same timescales globally. The UK appears ahead of global averages particularly in uptake of hybrid options and also depositing articles in OA repositories and websites. UK authors' uptake of Gold OA publishing options in particular rose by 65% between 2012 and 2014, moving from 12.6% of outputs in 2012 (slightly lower than the global average of 13.6%), to 18.2% in 2014 (above the global average of 16.6%). It seems reasonable to assume from this that the UK's approach to OA implementation is at least a contributory factor in a greater proportion of the literature being made available in an OA form. But the question is, at what cost?
In the context of this previous work, the research undertaken in the current study had the following objectives: The study therefore focuses initially on the APC data, discussing these in most detail as the newest large-scale cost area experienced by institutions, then goes on to discuss administration and subscription costs. These data are then brought together in the 'total cost of publication' modelling, with the question of the additionality of costs to institutions (a question at the heart of the 'double dipping debate) covered in particular. This provides the basis for a discussion of the policy implications of the research. At the same time, the study also identifies strengths and weaknesses of the datasets currently being collected, and makes recommendations on how data quality and the availability of data could be improved.

Methods
Data were assembled covering APCs, administration costs and subscriptions from several sources.
Data on expenditure on APCs for 2014 were collected with Jisc during the first quarter of 2015 from a sample of UK institutions. Jisc compiled the data into a single dataset. APC data were in two parts: detailed APC data (including a record of all individual APCs paid) from 24 volunteer HEIs usable in non-anonymized form comprising centrally-managed expenditure; anonymized data on 'headline' APC expenditure (including only total expenditure) from 23 the same HEIs covered in previous work. The headline data from the 23 institutions were added to data collected for our previous study in order to carry out a longitudinal analysis (reported below in the first section of the Results), whereas the 2014 data for the 24 nonanonymized institutions were used for the detailed analysis of the APC market (in the remaining sections of the Results below). Both datasets represent expenditure from institutions ranging from large research-led universities to smaller specialised institutions.
There was overlap between the 24 institutions in the 2014 APC data and the 23 in the previous study. They are reported separately here because of the agreement made with the 23 institutions when the first study was carried out that their anonymity would be preserved. Several issues arose in processing the data that are indicative of important aspects of the current APC market as experienced by HEIs. First, it was apparent that institutions were reporting some very low APC prices. These were normally explained by discounts often linked to pre-payment deals. For example, one institution recorded 40 APC payments made to a single publisher, Elsevier, averaging less than £40 each: this following a one-off deal with the publisher. There was also widespread use of schemes such as the Royal Society of Chemistry Gold4Gold scheme which resulted in some £0 being recorded (since subscribers were given vouchers enabling some APCs to be free). Such 'free' or highly-discounted APCs were normally part of wider deals with publishers (including some early offsetting arrangements) and so therefore need to be considered in this context of total costs to institutions (hence the importance of considering the 'total cost of publication', below, rather than just APC expenditure in isolation). These low costs were checked where possible and corrected (if an error was identified) or accepted (where a low or zero APC payment was verified). This research aimed to analyse what institutions were actually paying not simply list prices and, therefore, APCs were recorded at the discounted rate (not the list price).
Secondly, there was evidence of splitting of APC payments, normally between two funders.
For analysis, these payments were merged and the agency listed as paying the greater amount was recorded as the funder. For the few payments where there was an even split between funders, the first named funder was recorded.
Thirdly, some records of payments evidently included charges in addition to APCs. It was clear that colour and page charges were being recorded in the same payment details as APCs and were often apparent by anomalously high APC prices. Wherever possible, these were identified and excluded from the APC figures used for analysis. The extent to which the charges should in future be incorporated into 'total cost of publication' modelling is, however, a moot point. If data on these costs could be systematically assembled, there is a case for their inclusion in future analyses.
Fourthly, there was some inconsistency in the definition of 'publication date', i.e., between when the version of record (VoR) was made public on the journal web site and date when the VoR was made part of a volume and issue of a journal. There can sometimes be a considerable length of time between these two. However, it was impossible to correct this inconsistency reliably without wholesale checking of the data and it was therefore accepted as a feature of the data. Therefore, the APC dataset (now available in its 'cleansed' form on Figshare, Jubb, 2015), comes with caveats. Efforts were made to check and correct obvious anomalies but such efforts did not extend to verifying every single payment. It is likely, therefore, that the dataset still includes some inaccuracies and inconsistencies. Further work on standardising data collected from institutions is clearly required.
This research included only centrally-managed APC expenditure within institutions. HEIs are currently unable to report reliably on expenditure made elsewhere and so it is difficult to estimate levels of such expenditure. It is unlikely that payments of APCs outside the centre would occur at significant levels for RCUK or COAF-funded research or where institutional pre-payment schemes with publishers are in place, but they may occur for other research outputs, depending on institutional arrangements for funding of APCs (see below).
Analysis of the data was based on publication year as being the most easily publicly verifiable date but has the caveats outlined above. An alternative would have been to carry out analysis by date of payment, but there was insufficient data for both APCs and subscriptions to allow this. Data for calendar years, rather than financial year, were used as this is what was available for both APCs and subscriptions.
Administration cost data used were based on estimates from Johnson, Pinfield, & Fosci (2015), of an average of £88 per APC. This represents a total processing time of 134 minutes shared between faculty and support staff.
In addition to APC and administration data, subscription data used were already in the public domain (Lawson & Meghreblian, 2014b)

Centrally-managed APC expenditure over time
Previously, we reported a marked rise in centrally-managed APC payments from 2012 in the 23 sample HEIs . The new data (Figure 1) from the same institutions show this rise continued in 2014 approximately in line with, but higher than, the total projected (based on the three-months of data then available). In 2014, the same 23 HEIs spent £8,806,723 ($13,406,739) on centrally-managed APC payments. This rise is partly due to rising expenditure on APCs and partly a shift in existing levels of expenditure from distributed to centrally-managed budgets in HEIs. Since Jubb et al. (2015) calculate a 65% increase in paid-for Gold OA articles (compared with this rise of over 550%), it can be reasonably assumed that a large proportion of this rise is due to a shift in accounting in institutions from predominantly distributed payment of APCs to centrally-managed payments. At the same time, this shift has created much greater visibility of payments which are also likely to continue to increase in the next three years as compliance rates for RCUK and COAF-funded research outputs increase. The level of the increase may, however, vary between institutions, depending on local policies and payment methods, including whether they have also used money from other sources to pay APCs.  Payments ranged from zero (waived payments as part of deals with publishers) to £4,536 ($6,904) (mean=£1,586; $2,415). Where £0 payments were excluded (n=40), the mean was £1,599 ($2,435) (n=4,813).
Payments by institution are shown in Table 1 by institutional 'mission group': Russell Group (large research-intensive institutions), 'Pre-92' institutions (other research institutions), 'Post-92' institutions (teaching-led institutions) and 'Specialist' HEI. There were marked differences in numbers of payments made, from less than 10 (three institutions) to approaching 2,000 (UCL). These differences, in many respects, reflect the research intensity of the institutions and, therefore, the numbers of research outputs they produce. There were, however, evidently differences of policy between institutions which were reflected in the structure of their expenditure.   The data for funding source were, however, incomplete, with only 3,285 of the 4,853 records including a funding source. Of these, 2,152 (65% of those recorded) cited the funder as RCUK, 500 (15%) Wellcome, 249 (8%) COAF, and 288 (9%) internal institutional funds.
The remaining 3% were smaller amounts for a variety of funders. Centrally-managed payments were, therefore, largely being generated by external grants designed specifically to fund APCs with internal funds being used less commonly. With funder preference for licences that allow for liberal reuse (including commercial exploitation), it is unsurprising that 89% (1,909) of APC records in the dataset with the licence field completed (2,146 (44%) of the 4,853 total) were listed as having a CC BY licence (one of the Creative Commons licences allowing most extensive reuse).

Figure 4: The range of APC prices paid by institution for articles published in 2014
The journals for which APC payments were made were mapped against the broad subject panels used in the 2014 REF using subject classifications from Scopus in order to assess their broad disciplinary coverage (   Most publishers charged a relatively wide range of different APC prices. Figure 5 illustrates the price range of APC payments for the top-10 publishers. It is noticeable that payments to Nature and Elsevier cover a wide range, including very low levels for Elsevier due to one-off discounts included in deals with HEIs. There is also a marked difference in the median price among the different publishers. Two publishers had median APCs below £1,000: Frontiers, £902 ($1,373) and PLOS, £972 ($1,479). Two publishers had median APC levels above £2,000: OUP, £2,100 ($3,195) and Nature, £3,360 ($5,109).

Figure 5: Range of APC payments for the top-10 publishers measured by receipt of APC payments
Analysis of the APC expenditure by journal type shows a marked difference between the mean APC charged by hybrid journals and OA journals, with hybrids considerably more expensive (Table 4), consistent with previous studies (Björk & Solomon, 2014a, 2014bPinfield et al., 2015). There is also a difference between fully-OA journals produced by publishers who also publish subscription titles and those who publish only fully-OA titles (previously observed by Björk & Solomon, 2014a, 2014b. The hybrid mean APC is 58% higher than the mean of fully-OA journals from 'non-subscription' publishers. However, journals may offer different levels of service and may deliver different products (most hybrids, for example, deliver their product in paper and electronic format, whereas fully-OA journals do not). These points (as well as price) need to be considered in any holistic comparison. Nevertheless, the price differentials are considerable.  An interesting question arising from this relates to the relationship between price and quality.

Number of journals
To address this, the APC price data were matched against Field Weighted Citation Index (FWCI) scores derived from Scopus to test whether there was a correlation between APC price and citation impact, using citation impact as a proxy measure of quality. Initial analysis of the journal types in Table 4 shows a correlation between price and citation impact ("Ave. FWCI" column). To test this further, journals were grouped in 10 different FWCI categories for analysis, with all journals covered in Scopus being ordered according to their FWCI and then ranked into tiers, each tier accounting for 10% of the total number of journals, the top tier rated 1 and the bottom tier rated 10. To provide greater granularity the top level, which accounted for 38% of articles, was further divided in two, with the top 5% rated 1 and second 5% rated 1.5, making a total of 11 tiers. For each tier, Table 5 shows the numbers of journals and of articles for which APCs were paid from the sample. The proportions of those journals and articles for the whole sample are also given.  There was a strong correlation between APC price and FWCI ( Figure 6): 90.4% of the variation in mean APC was explained by mean FWCI. This is consistent with another recent study of APC list prices (Björk & Solomon, 2015) which also found that highly-cited journals charge higher APCs using different citation indexes and based on list prices. Highly-cited journals charging higher APCs may, of course, be explained in different ways. High-FWCI titles tend to be more costly to produce (with, for example, higher rejection rates and more rigorous editorial standards): higher APC prices may therefore reflect higher costs.
Conversely, authors clearly value publication in highly-cited titles more and may be prepared to pay higher APCs. Higher APCs may, therefore, reflect the fact that the market will stand higher prices. It is possible that both of these factors may be important. Further work could investigate this correlation, including more detailed comparisons of hybrid journals and fully-OA journals with similar FWCI scores, and further work could examine value and cost (not just price). Based on Scopus data, the number of UK OA articles for which an APC was paid rose from the 15,444 in 2012 to 25,001, a rise of 65% . Our sample of 24 universities produced 4,853 articles in 2014 for which centrally-managed APCs were paid, part of steep rise in centrally-managed payments (as reported above). However, based on estimates derived from Scopus of the total paid-for Gold outputs of the institution, we estimate that APCs paid from centrally-managed budgets rose from 20% of the estimated total APCs paid by the institution (747 of the 3,786 Gold articles) to 78% in 2014 (4,853 of the 6,250 Gold articles). This is a considerable shift in favour of centrally-managed funds but one that shows that 2014 data presented here is likely to represent a large proportion of the overall payments made by HE concerned, with non-centrally managed payments being 22% of the total.
Two of the Russell Group institutions from our sample also provided estimates of expenditure from non-centrally-managed budgets based on an analysis of expenditure recorded in their institutional financial system. Both institutions identified records of APC expenditure in their institution outside the centrally-managed funds for a sample of publishers corresponding to the top-10 publishers identified by this study (Table 3). One institution reported that total expenditure from non-centrally-managed budgets was as high as 31% of the whole, whereas the other reported just 14%. Such differences may be due to different institutional policies and varying publicity about the availability of centrally-managed funds.
It is interesting that the second institution with a lower level of expenditure from noncentrally-managed funds is one where internal institutional funding has been added to the central fund allowing authors to pay APCs even where they do not have a grant from an external funding body. This is not the case for the first institution and may mean that users there have less opportunity to use the central fund.
These estimates compare with those made by Pinfield & Middleton (2016)  What is clear, however, is that current datasets of centrally-managed APC payments under-record total payments made by the institution as a whole, a fact which needs to be taken into consideration in the 'total cost of publication' modelling below.
Anecdotal evidence indicates that well-established fully-OA publishers that do not have prepayment deals in place with institutions may receive a much greater number of non-centrallymanaged payments than hybrid publishers. Fully-OA journals may have established relationships with authors and the comparatively low prices charged for APCs mean that authors are willing to pay from local funding sources. Hybrids, on the other hand, tend to be paid centrally to a greater degree. However, such observations are very impressionistic and need further testing. In contrast, it is clear that other costs of publication, e.g., colour and page charges, might also arguably be included in the total cost of publication: these tend to be distributed around institutions and are therefore often difficult to identify (Gray, 2015).

Subscription expenditure
Subscription data for 24 UK higher education institutions were available in useable form for  In addition, in most cases, there was also a year-on-year increase in the aggregated subscriptions paid to publishers (Table 6)    in which online access is predominant and site licenses the norm, it is unlikely that other subscriptions exist to any large extent in the institution apart from those paid centrally. This is different from APC data which, for reasons outlined, are likely to under-represent institutionwide cost.

'Total cost of publication'
Previously, we used the measure of the 'total cost of publication' (after Willetts, 2014) to gauge the new additional costs being experienced by institutions . The measure consists of the cost of existing subscriptions plus APCs and new administration costs, and since it was first used in the context of trying to develop an evidence base relating to perceived 'double dipping' by publishers (and also concerns about rising additional administrative costs), the measure only includes hybrid journals since only these involve a business model with two income streams (subscriptions and APCs) from the same customer for publisher content. Fully-OA journals might more reasonably be seen as an alternative cost to subscriptions. Use of only hybrid APCs for the modelling may also be preferable from a pragmatic viewpoint, since if it is believed that a larger proportion of fully-OA APCs are paid outside the centre than hybrids, then the data for hybrids are more likely to be a closer reflection of the overall institutional payments. The data now available mean that the 'total cost of publication' can be refined and updated using the current APC data in combination with subscriptions data. These can be combined with previously-calculated administration cost data of £88 per APC (Johnson et al. 2015).
The calculations are shown in Table 7 and Figure 8. Across the 24 institutions and seven publishers, subscriptions constituted 87% of the TCP, APCs 12% and administration costs, less than 1%. These show a slightly higher proportion of costs in the APC category than previously (Johnson et al., 2015;Pinfield et al., 2015) but are not directly comparable, since their study covered all subscriptions paid to publishers to which APCs had also been paid, and was, therefore wider in its coverage.  Since it has been shown that subscription expenditure has risen for institutions in the last five years, it is clear that APC costs shown here are largely additional costs for institutions, at least currently. There is no evidence in the data examined that subscriptions have declined commensurately as APC expenditure has increased. Moreover, APCs have been included here at discounted rates, if applicable, and therefore take into account at least some of the offsetting that occurred. The additionality of APC expenditure is also apparent because whilst subscription costs given can reasonably be assumed to represent the entirety of an institutions expenditure, the APC data under-represent actual expenditure (as discussed above). The evidence therefore indicates that, currently, the APCs in the calculations here are additional costs, a situation that is likely to continue for the foreseeable future until any institution-level offsetting agreements are more widely adopted and have had time to take effect.

Discussion
Addressing the question of the cost of Gold OA requires multiple strands of evidence and a The key issue facing policymakers in particular is whether these additional costs currently faced by institutions might be considered transitional or whether they will remain in place in the long-term. For them to be considered transition costs, there needs to be a clear 'line of sight' between the current subscription/hybrid OA system and a situation where OA becomes the predominant model of scholarly communication. In practical terms, this would mean evidence of publishers transitioning their business models to incorporate some kind of 'offsetting' and, ultimately, 'flipping' arrangements (i.e. replacing subscription-based with OA-based models). In the current market, publishers are developing new business models which, in some cases, involve offsetting arrangements, including ones which appear to allow for flipping within the foreseeable future (Springer, 2015). Where this is the case, however, the complexity of the different publishing models and a variety between them can create challenges for institutions. A more fundamental challenge for HEIs is that not all publishers have accepted the transition assumption of HEIs or have meaningful offsetting arrangements or plans for flipping in place.
Our study provides empirical evidence that some offsetting is occurring, particularly as discounted APCs for subscribers. However, the additionality of APC costs indicated would suggest that, at present, any offsetting taking place does not fully compensate for the additional costs of Gold (including hybrid) OA. The need from an institutional point of view for HEIs (or consortia) negotiating with publishers to build an understanding of the 'total cost of publication' into the negotiations, resulting in meaningful offsetting, is becoming more apparent. Significantly, this is reflected in the UK by the recent statement of principles articulated by Jisc on offsetting (Jisc, 2015), which builds on policy positions of UK funders to engage with hybrid as a transition approach (Finch et al., 2012;Jubb, 2014). Important features of this include the aim that offsetting models lead to benefits for specific institutions with greater APC expenditure (rather than just based on global averages) (Björk & Solomon, 2014a), and should involve ongoing offset business processes which can be reasonably administered by the different stakeholders. On the last point, it is clear from our study that administrative costs are currently only a relatively small proportion of the 'total cost of publication', but their place in the 'total cost of publication' to be monitored along with other costs associated with OA management.
The success of HEIs in the UK and elsewhere in negotiating offsetting arrangements could impact more widely on the perceived future viability of hybrid OA. The recent Max Planck Society initiative calls for international action around offsetting and flipping, identifying offsetting agreements as "the most promising" options for achieving a transition from subscriptions to fully-OA publishing (Schimmer, Geschuhn, & Vogler, 2015). However, apart from libraries demanding offsetting deals "more energetically" and with greater international coordination, it does not suggest mechanisms for achieving such a transition.
Apart from engaging in negotiations, as costs rise, there are, of course, various other options open to HEIs. Funding agencies may withdraw from or limit funding for hybrid journal APC payments (either by capping the level of APC supported or limiting the total budget available for hybrids). However, the evidence for the UK suggests this would immediately reduce the proportion of outputs available in OA. The negotiation of a cost-neutral transition to OA may, on the other hand, be difficult to achieve, at least in the short term. The question in that case will be how long is it acceptable for additional transitional costs to be borne by funders and institutions? One possible compromise for funders and institutions keen to encourage hybrid as a transitional mechanism might be to pay APCs for hybrid journals only if the publisher has in place an offset model which has been in some way 'approved' by the research funder.
This would, of course, necessitate the production of clear criteria for such approval, but documents such as the Jisc offset principles could be a good basis for such critera (Jisc, 2015). Such an approach might itself encourage offsetting agreements. A related approach might be to establish criteria for the value of various OA-related services publishers' provide (licenses available, deposit in repositories etc.) and fund them according to these criteria. This would, of course, require detailed work to establish relevant cost and value criteria. In all cases, it will be important to continue to develop an evidence base to inform such work both in the UK and internationally. The data examined in this study was of a better quality and more easily available than that reported in Pinfield et al. (2015); however, there is still a pressing need in the UK and internationally to ensure that more institutional data, collected and shared in more standardised ways are available to inform ongoing developments.

Conclusion
This study demonstrates the APC market is currently complex, with variable pricing, discounts, and other additional charges contributing to institutional costs. Nevertheless, it is clear that, in the UK, centrally-managed managed APC expenditure has continued to rise steeply. This can largely be attributed to the policy and funding position amongst UK funding agencies. APC payments in 2014 varied from £0 to £4,536 ($6,904) (mean=£1,586; $2,415).
There was considerable variation in the levels of payments across different institutions, reflecting research activity and policy differences. The largest number of institutional APC payments was made for articles in the Health and Life Sciences. Commercial subscription publishers were responsible for the largest proportion of the centrally-managed APC market in HEIs, many offering hybrid options. Hybrid options were, nevertheless, considerably more expensive than those for fully-open access titles. However, there was a correlation between APC price and journal quality (shown in journal citation rates).
APCs in the 24 UK institutions (using a sample of seven publishers) now constitute 12% of the 'total cost of publication' with APC administration, 1%, and subscriptions, 87%. This is at a time when subscription costs for institutions have risen, indicating that APCs and administrative costs currently constitute additional costs for HEIs (i.e. subscriptions have not declined commensurately as APC costs have increased). There is likely to be an impetus to review current Gold-centric policy positons and funding arrangements in the UK. Whilst the approach appears to have resulted in an increase in take-up of OA, it has created major cost pressures -pressures illustrated in the 'total cost of publication' modelling. Time will tell how these pressures will be addressed in terms of policy development, both in terms of continued pursuit of Gold OA and the incorporation of Green OA in policy approaches.
The developing evidence base can inform policy development in the UK and internationally about the shape a Gold-centric approach can take, including the extent to which support for hybrid journals is a transition mechanism. The pattern of market activity of HEIs in the context of the UK's largely Gold-centric position to date and the challenges it has created, along with discussion of the ways in which those challenges may be addressed in future, will have a significant impact on ongoing development both within and beyond the UK.