Collection of text and image files from WWW galleries and exhibits took place over a five-month period, from April 7-September 7, 1994. Galleries were located by keyword searches of various Web search engines including the CUI WWW catalog and NorthStar, and by following hypertextual links from art resource lists compiled by Anima, Fine Art Forum, and other collectives serving the online arts community. Institutional galleries (e.g., museums), exhibits from visualization laboratories, and mere archives (uncurated collections of unattributed gif, jpeg, and mpeg files) were excluded from the investigation. An effort was made to achieve a representative sampling of sites with respect to such factors as the number of exhibitors and the presence or absence of commercial interests. The group of thirty-three sites selected included sixteen exhibiting the work of a single artist, and seventeen in which two or more artists were represented. Fourteen of the galleries were hosted by commercial service providers, one was available from a server running on a "home" computer, and eighteen were housed on servers at educational institutions. Nine of the thirty-three galleries offered works of art for sale. (Although the population of commercial galleries on the Web is growing, their number is quite limited at present.) Finally, an attempt was made to ensure that the galleries selected offered a wide range of both electronic and traditional works of art. The galleries and exhibits included in the study were the commercial sites The 911 Gallery, Access Art, the Anecdote Gallery, Cloud Gallery, The Electric Gallery, Kaleidospace, the Paulette Jellinek Art Gallery, the Pearl Street Gallery, and the Pixel Pushers Electronic Art Gallery, and an additional group of sites not offering works of art for sale: the Amiga Art Gallery, the Anecestry, Religion, Death and Culture exhibit, Beauty for Ashes, Chez Rampart, Cirque de la Mama, David Miller exhibit (dem's pictures on the Web), the Generality galleries Drux and aart, the USC Ethnographics Laboratory Exhibit, Filter Art, the Fine Art Forum Gallery, Fractal Pictures and Animations, the Gate Gallery, The Hall of Self-Expression (Margaret Watson gallery), Henry Houk's Photo Gallery, In the Pockets of the Night (Elizabeth Heron exhibit), International Interactive Genetic Art, the Jane Patterson Art Gallery, Marius Watz' Art Gallery, NEXUS, Off the Wall Gallery, OTIS, Strange Interactions (John Jacobsen exhibit), Avi Rosen exhibit, and Studio X: MkzdK. Each of the 33 sites was viewed at least twice during the data collection period. Exhibits were viewed using Mosaic or Cello over a SLIP connection to a local commercial service provider; thus the galleries were experienced via the comparatively slow serial connection typical of the average Web user rather than a fast network connection. From each site all text files and most image files were retrieved, and most were saved, in their original format (html, gif, jpeg, mpeg. etc) for subsequent viewing and analysis. At the smaller sites, all of the images were saved; at the larger (> 30 artists) a randomly selected image for each exhibitng artist was retrieved and saved. Some sites presented consistent problems with respect to retrieval of files, some had sections still under construction, and others had lapses in their indexing which sometimes made it difficult, if not impossible, to determine the name of the artist responsible for a particular work. Because the data were collected over a five-month period, the data corpus may not reflect the work of very recently added artists in some of the largest of the galleries. The nature of the hypertextual medium did in some cases make it dificult to determine the "walls" or boundaries of a gallery. In ambiguous cases the rule of thumb was that the home page was that which contained the gallery logo and direct links to the artworks. Only information that could be accessed directly from the home page, and the documents to which the home page pointed, was included in the data corpus. Thus, for example, data was not included which could only be obtained by bringing up and searching the home page's parent directory The data corpus of texts (gallery mission statements, artists' biographies and exhibition records, image captions, hotlists) and images and animations (logos, fancy buttons, imagemaps, menu bars, color bars, artists' portraits, works in the collection) was reviewed to generate a preliminary account of gallery characteristics. This preliminary review led to the development of a list of features which represents the wide range of exhibition practices and possibilities on the Web. The features were grouped into three broad categories, which can be regarded as dimensions or continua of electronic exhibition space along which, separately or conjointly, exhibitors may vary. The first of these three dimensions is network embeddedness: the extent to which a site is central to, or visible in, the matrix of online relationships among galleries, their viewers, and their contributors. Network embeddedness may be reflected in indices of pointing and cross-pointing (mutual hypertextual links between sites); in the presence of mediated extensions of the gallery's outreach, such as affiliated gopher or ftp sites, mailing lists and newsletters; and in university or studio affiliations. The motivation to achieve network centrality may be indicated by features such as contests and awards, giveaway or sale of promotional items such as t-shirts or posters, vanity publishing of photos and comments, and sponsorship of online meetings, face-to-face meetings, and art-related special events such as symposia or tutorials. The following three items were used as the feature list for the measurement of network embeddedness: 1. Number of other surveyed art sites pointing to the gallery 2. Number of other surveyed art sites pointed to by the gallery 3. Number of networking/promotion strategies employed, such as tips for artists, vendor presence, newsletter, mailing list, resources list, promotional items, affiliated newsgoup, sponsorship of collaborations, on-line meetings, posting of annotations, professional organization / arts collective link, affiliated gopher/ftp site, resident columnist/artist, bibliographies, databases/archives, contests and awards, criticism/ reviews, discounts/specials on art-related materials, affiliated teaching/ academic program, affiliation with/sponsorship of events (festivals, seminars), studio/design firm affiliation, associated publication, reference to funding, public or private support, vanity publishing (will post your photo, comments, etc), and presence at professional meetings WWW galleries and exhibit spaces may also differ with respect to their openness to visitors and to potential exhibiting artists: the extent to which information about the gallery, its exhibitors, its art objects, and the mechanics of retrieval and viewing are available freely and in a manner friendly to vistors with varying degrees of online literacy. Openness also refers to the extent to which the gallery allows, encourages, or solicits submissions and offers a variety of means, both traditional and electronic, of submitting works for exhibition. Openness may be reflected as well in the extent to which the exhibit space embraces the visual arts in all their variety. The following three items were used to operationalize openness: l. Informativeness: amount of information about the artists and the images/animations, including the presence of any of the following: artists' name, artists' portraits, artists' bios, resumes and/or exhibition records, artists' personal statements, artists' home pages, hyperlinks on artists' statements, artists' email addresses, artists' surface mail addresses and/or phone/FAX numbers, pictures of the artist's studio/shop; and information about the works of art, including title/caption, extended caption, date produced, approximate file size, image format, price, dimensions, how work was made, number of prints in the series, and physical location of the collection 2. Visitor-friendliness: Number of measures implemented to encourage viewing and contributing by users varying widely with respect to technical knowledge and access, including invitation to submit/ information about how to submit, submission methods for non-computer- literate artists (surface mail diskette, unscanned photo), choice of file format (compression, size, resolution), links to viewers , browsers and graphics software and information, and index or catalog 3. Openness to a variety of forms of the visual arts; number of different types of visual art displayed, such as digital images (computer- generated or manipulated images), including digital photographs, works created with paint or photomanipulation programs, rendered works (raytraces, 3d), and algorithmic art (fractal images, genetic art); analog images (digitized reproductions of artworks created with traditional media, including photographs, drawings (pen and ink, pencil, charcoal, chalk), paintings (oil, acrylic, watercolor, tempera), prints (woodcut, engraving, lithograph, silk-screen), and other miscellaneous artworks ( mixed media works, CD covers, rave flyers, sculptures and assemblages, installations, textiles, furniture, glass, ceramics, clips from videos); moving images, including animations (fractal animations, genetic algorithm animations, morphs, raytrace and 3d animations), and digitized videos; and other art forms with images, including graphic novels, comic books, illustrated poems, essays, and short stories
Web galleries and exhibits may also be seen to differ with respect to the design concept which informs them: the mix of the artistic and tehnical sensibilities which determines the look and 'feel' of the space. The design concept is reflected not only in the way in which the visitor is "walked" through the exhibit space (through "rooms," "wings", or past a "wall space"), but also in the use of artwork in logos, fancy bullets, and inline images, and more technically advanced use of artworks in hypertextual features such as fancy buttons, menu bars, and imagemaps. The design concept is futher reflected in the presence of interactive features, which organize the activities of the gallery through forms which allow visitors to send comments to the curator or to the artists, search archives or databases, or order objects on display at the site. Features that were used to characterize the galleries with respect to the dimension design concept were:
1. page design artwork, including: original logo, fancy bullets, legend, inline images, color bar, thumbnails (interactive), fancy buttons, menu bar, imagemap, and player bar for movies
2. advanced interactive features (forms), including comments page (email to curator, artists; guest book), keyword search of archives or database, order form, submit form (work, resources, links), voting form, and questionnaire/survey form
3. strategies for organizing the collection-- by artist, by medium (image/ animation; painting/drawing/print), by contexts (geographic, chronological, style), by analogy to physical spaces (walk, room, wing, annex), and by individual works
Other characteristics of the exhibit space included in the analysis reported here are the number of artists exhibiting at the site, and proportion of exhibiting artists who were female. Gender was estimated by a "best- guess" method based on gender-linked names and by the information provided in artists' biographies. Additional characteristics recorded, but not included in the analysis reported here, were availability of information about the curator/webmaster; the range of file formats supported; strategies used by the gallery to protect the copyright of the artist (for example low- resolution or altered versions of the images); and any possible sources of difficulty for the site (nudity in some of the works in the collection, persistent difficulties in retrieving files from the site). Data recorded for the commercial sites included the type of items for sale (originals, artist's proofs, limited edition prints, open edition prints, reproductions), the method of ordering (email, surface mail, phone, FAX, fill-in form), the method of payment, the availability of secure means for the transport of credit card information, the availability of studio visits, willingness to provide slides for previewing, and various guarantees related to authenticity of the work and satisfaction with the purchase. Analysis of the Data Variables in the analysis described below included two grouping variables, single/multiple exhibitors and commercial/ noncommercial gallery. Dependent measures were number of exhibiting artists, proportion of exhibiting artists who were women, network embeddedness, openness, and design concept. Alpha reliabilities were computed for network embeddedness, openness, and design concept. All were satisfactory: standardized item alphas were, respectively, network embeddedness, .83; openness, .78; and design concept, .78. The variable measuring strategies for organizing the collection had a poor item-total correlation and was dropped from the design concept scale. Cluster analysis The first analysis of the data was designed to provide a general overview or topography of gallery-space on the Web. Using only the dependent measures, and not the grouping variables, observations of thirty-two galleries on the five variables were subjected to a cluster analysis based on the technique of nearest centroid sorting (SPSS QUICK CLUSTER (Norusis, 1990)). (One of the galleries had to be dropped from the analysis because the number and gender of the artists could not be determined.) In the clustering method used for the first analysis, a case is assigned to that cluster for which its distance from the cluster centroid is least. The analysis was performed three times, assuming either a two-, three-, or four-dimensional cluster solution. ANOVA was used to test the ability of the dependent measures to discriminate among the cases by cluster membership. The four-cluster solution was chosen because the ratio of between-to-within cluster variability was significant for four of the five dependent measures: number of artists, F = 45.76, p < .0001; proportion of women exhibiting, F = 101.79, p < .0001; network embeddedness, F = 26.25, p < .0001; openness, F = 15.30, p < .0001; design concept, F = .5568, ns. The table below provides the means of the four clusters on the five dependent measures. Table 1: Means of Four Gallery Clusters on Dependent Measures Cluster #Artists %Women Embeddedness 1 5.3121 10.9084 2.5023 2 99.0000 6.0000 28.0000 3 5.6548 27.2308 33.7302 4 1.0476 99.2326 1.2284 Cluster Openness Design Concept 1 7.2776 2.9465 2 29.0000 2.0000 3 16.9680 3.6307 4 8.2312 1.8943 Cluster 4 contained nine galleries, all but one of which were one- woman shows. Cluster 4 member galleries were characterized by low network embeddedness, less informativeness, and less sophisticated design concept. Cluster 3 consisted of three galleries with a small but eclectic mix of artists, of both genders. Cluster 3 members were characterized by strong network embeddedness, moderate openness and an advanced design concept. Cluster 2 consisted of a single gallery with a roster of exhibiting artists far in excess of the average, in part due to its long (by net standards) tenure as an ftp archive prior to its debut on the Web. The Cluster 2 gallery is characterized by its superior network embeddedness and openness, relatively less sophisticated design concept, and by a comparative absence of female exhibitors. Cluster 1 consisted of the remaining galleries, which are small-to-medium-sized sites with low embeddedness, low openness, few women, but average to above average use of forms and artwork in their design concept. It will be noted in Table 1 that the variability of the network embed- dedness concept is greater than that of the other dimensions, taken across all clusters. It is interesting to note that in the sample of 33 galleries, there were in total only 77 instances of pointing (providing a hypertextual link) to other galleries included in the list. Of these 77 instances, only 12 of them were associated with cross-pointing, or mutual hyperlinking between pairs of galleries. Five of the 33 galleries were complete isolates within this group of sites: they neither pointed nor were they pointed to. For-profit sites were far less likely to use pointers than noncommercial sites; in fact, only one of the commercial galleries observed pointed to the other galleries. Although the number and gender of the artists featured prominently in the cluster analysis, the commercial/noncommercial distinction was not a particularly strong discriminator; commercial galleries did not appear to cluster together, but in fact were scattered across three of the clusters. In the next section the results of a discriminant analysis designed to assess distinguishing characteristics of the commercial sites is reported, along with a further examination of the differences between single- and multi-exhibitor galleries. Discriminant analyses Discriminant analysis provides a multivariate-analysis of-variance test of the hypothesis that two or more groups differ significantly on a linear combination of a set of dependent measures. To assess what features if any distinguish commercial from non- commercial galleries, a stepwise discriminant analysis was conducted in which the dependent measures were number of artists, proportion of artists who were women, network embeddedness, openness, and design concept. One of the thirty-three cases had missing values on the number of artists and gender measures and was consequently dropped from further analysis. The Wilks' lambda criterion was used for the discriminant analysis, with values of F-to-enter and F-to-remove set to 1.0. Prior probablities were set at the default value of .50. Box's M was 7.97 (p < .34), indicating that the within-group covariance matrices were homogeneous. The discriminant analysis resulted in a linear combination of three of the dependent variables, network embeddedness, openness, and design concept, which maximally discriminated between commercial and noncommercial sites (lambda =.75, Chi-square = 8.172, p < .04). The standardized discriminant function coefficients for the three variables were, respectively, network embeddedness, -1.01; openness, .78, and design concept, .76. The commercial galleries group had the higher mean score on the discriminant function, indicating that commercial sites generally were less embedded but more open and more advanced in their design concept than the non-commercial galleries. A test of the predictive ability of the discriminant function indicated that the linear combination of the three variables was able to reclassify 82% of the galleries correctly with respect to the commercial/non-commercial variable. Univariate comparisons of group means indicated that the commercial galleries also had marginally more artists and a greater proportion of women; these differences were not statistically significant. Findings of significant differences must be interpreted with caution in light of the small size of the sample of commercial galleries. A second discriminant analysis was undertaken to probe further the differences between single-artist exhibits and multiple-artist galleries. Dependent measures included in the analysis were proportion of women artists, network embeddedness, openness, and design concept. The discriminant analysis produced a linear combination of three of the variables, proportion of women, openness, and design concept, which significantly discriminated between single-artist and multi-artist sites (Wilks'lambda = .48, Chi-square = 20.81, p < .0001). Standardized discriminant function coefficients were, respectively, proportion of women, -.59; openness, .69; and design concept, .79. The group containing galleries with two or more artists had the higher mean score on the discriminant function. Single-artist sites had a higher proportion of women exhibitors, less network embeddedness, less openness, and a less sophisticated design concept. Univariate tests indicated significant differences on all of the variables but openness, which just missed significance at p <.07. Results of reclassification based on the discriminant function indicated a success rate of 87.50%. The finding of significant differences must be interpreted in light of a demonstrable lack of homogeneity in the group covariance matrices (Box's M = 17.5, p < .02) and the small sample size. To summarize briefly, art sites on the WWW can be grouped into broad clusters or types on the basis of their size with respect to the number of artists and their composition with respect to the relative number of female artists. Both gallery size and the raison d'etre (commmercial/noncommercial) of the exhibitors shape the form the gallery ultimately takes with respect to such factors as network centrality, openness and sophistication of gallery design. Commercial sites, at this early stage in the development of the Web, seem to have the advantage in the innovative application of forms and advanced design strategies. This reflects not only the ability of the commercial gallery principals to hire professional design teams, but more fundamentally the greater need of such undertakings to develop a customer base and offer security and convenience in their business transactions. Commercial galleries also appear to be marginally more open to the display of art works by women. This appears to be an encouraging development, as women artists on the net are far more likely than male artists, relative to their numbers, to put on their own, one-person shows rather than participate in the larger galleries with their superior opportunities for networking and promotion. As we have seen, the one-artist sites were consistently disadvantaged with respect to gallery design and network centrality. This report has only touched on a few of the aspects of online exhibition culture. Other topics of interest include copyright issues and the prospects of censorship and control for artists exhibiting via commercial service providers.