Workshop Presenters & Attendees,

Welcome, everyone to the discussion thread for panel 5 “Using What Middletown Read” and thank you to Lynne, Doug, Steve, and Alex for two fascinating and engaging talks. As the discussant for this panel, I have the opportunity to ask the first questions to get this discussion going.

Lynne, Doug, and Steve: Did portrayals of the North factor into this southern corpus and if so, how did that play out in borrower activity? It was interesting that New York and Washington were the 2nd and 3rd most common states listed in the Southern Corpus. Additionally, Kentucky represented nearly 9% of mentions, over 3x as many as any other southern state outside of Virginia. Why was it so common and do you see that as a significant element in the diffusion of the Lost Cause myth to readers?

Alex: I am intrigued by the 3rd factor in your key factors to improve the model (“Authors who wrote in a particular genre, movement, or otherwise relatively niche part of the literary marketplace”). What do the results of your model have to say about genres more specifically? Did you find certain genres more apparent in borrower patterns and predictability? Finally, what impact did the demographics of the borrower play in the accuracy of the model?

Both: What is something that surprised you when working on these projects? Was there something that maybe you hadn’t thought about or challenged an assumption?

I would invite other attendees to post their questions and participate in a discussion of these two very promising projects.

Great questions, Jordon. We didn't think about representations of other regions of the United States, except for the South and in our conclusion the West. But it's quite possible to imagine doing so, and also to think generally about representations of other places (Germany in the Wister translations especially). It would be interesting to know if, like representations of the South, representations of the North are as variable, ranging from the substantive (e.g., in House Behind the Cedars) to the shallow (as in the romance-first books like Lena Rivers).

We don't have a tested hypothesis for the prevalence of Washington and New York. Part of the prevalence of New York is due to place-of-publication information (title pages, etc.) in the plain text, although the texts contain plenty of other mentions. It would be interesting to see a map plotting place in the Southern corpus vs place in a larger subset of Muncie-circulated fiction. I'd be willing to hypothesize that Washington and New York figure large in the geographical imagination of fiction in the library. If we rework the article for publication, I'll be sure to lobby for the testing of such a hypothesis.

~40% of Kentucky references come from the books of James Allen included in our corpus. I'd like to hear from someone who understands regional literature in the late 19th century. What counts as a "region"? We know there are Indiana-related books in the library; does Kentucky count as a kind of "almost Indiana?"

@jfbratt  Thanks for the thoughtful questions, Jordan.

@steve_pentecost answered the primary question best. I can toss a couple tangential observations into the conversation.

First, characters in fiction of course don't always stay put in a fixed geography. When novels with Southern themes deal with the Civil War, for example, there are naturally often references to Northern places and to Washington, D.C., not least in generally pro-Union adventure novels for boys. Oliver Optic's Within the Enemy's Lines, for example, involves cousins and uncles on both sides of the war and improbably heroic deployment of a family yacht that ranges far from its home base on the Hudson.

Regarding regionalism, one interesting thing that came out of topic modeling is that at a certain point we discovered dialect terms split into two topics, a stereotypical Black dialect topic and a second topic that looked like unmarked white Midwestern dialect. Curiosity about this led me to look briefly into poetry rather than fiction.

Poetry wasn't especially popular in Muncie compared to fiction. But within poetry, Indiana's own James Whitcomb Riley seems to be the most-circulated poet in the Muncie records, with one or another of his books usually in circulation every week or so. Riley achieved fame writing and performing Midwestern character-dialect pieces. Will Carleton's Farm Ballads and City Ballads were also in the library and had a few dozen circulations each. John Hay's Pike County Ballads, also known for representing Midwestern dialect, was not in the library.

Especially interesting in this regard are the two volumes of Paul Laurence Dunbar's poetry that were in the Muncie collection, and that seem to have been relatively popular. Lyrics of Lowly Life circulated 73 times after coming into the library in November 1898. It looks to me like it might be the one of the most-circulated volumes of poetry in the collection.

Dunbar's relation to dialect is complicated. William Dean Howells singled out the originality of his Black dialect pieces in his introduction to Lyrics of Lowly Life, but if you look at the contents of the volume, many or maybe most poems in the collection are not in that kind of dialect, but in standard 19th-century literary-poetic English. Dunbar certainly did include plainly Black dialect poems, but he also published Midwestern dialect pieces that sound a lot like Riley or Hay. Dunbar's "The Old Apple Tree" seems meant deliberately to evoke Riley's "The Old Swimmin' Hole."

Dunbar, from neighboring Ohio, showed facility with a variety of American registers and voices, but his public reputation seems to have been constrained by the more limited desires of white audiences for a certain representation of Black dialect. One can wonder how Dunbar's volumes might have been read in Muncie in the 1890s when the volumes were new.

@douglasknox @stevepentecost @alexleslie

There's a conversation to be had here around regionalism that builds on Jordan's question.  Alex's presentation on this panel shows us how the WMR data can be analyzed using logistic regressions (and overlaps in interesting ways with what Robin Burke and John Shanahan are up to with the Reading Chicago Reading project, discussed in panel 4).  But his dissertation focuses on how to look at regionalism, challenging prevailing claims about that genre's ideological significance.  I've dipped into the issue of regionalism and reading a little bit as well, as part of a special issue of the Middle West Review (7:1 (Fall 2020) that has several entries that might be of interest.  And of course, Christine Pawley's Osage study emphasizes regionalism as well.

One thing that seems to matter in all of these instances is that the social and geographic contexts in which reading of regionalism happened had significant influence on reception. 

There are also overlaps in methods between Alex's broader work and that of Doug, Steve and Lynne that are worth discussing further. 

One other note for Lynne, Steve, and Doug: 

Barbara Hochfeld's Uncle Tom's Cabin and the Reading Revolution (2011) argues that late nineteenth-century readers understood Uncle Tom's Cabin much differently than antebellum readers had.  She contends that the 1890s reading of Stowe's book was more in line with the kind of reading that you describe, reinforcing racism and romanticizing plantation life.

The MPL acquired a copy in 1891 and circulated steadily, consistent with Hochman's argument. I didn't see it cited in your paper, so I wanted to take note of it especially since Hentz's The Planter's Northern Bride figures so heavily in your analysis.

An impressive effort, Alex. Dazzling, in fact. I'm going back to watch it again, as soon as I catch up with the other presentations. If I may, I'd like to ask two or three questions:

To what extent does uncertainty figure twice in the model? Take, for example, the ideas a) that new books are often checked out quite often, and b) that a checked out book can't be checked out until returned, which would seem to limit the exercise of taste. If I understand that model in abstract terms, I'd conclude that a ("new books") figures into taste, and that b ("a checked out book . . . ") would count as count as some sort of error.

Does the regression function you're using provide a way of understanding which features are most informative, either overall or on an observation-by-observation basis? I'm not sure this is an entirely reasonable question, but thought I'd ask . . .

I'm also interested in Christine Pawley's talk within the context of yours. If I understand the implications of her talk, libraries also function as limits on taste, as a way of protecting readers from perhaps harmful materials and of directing them in perhaps more edifying directions. Have you thought about how this influences what you're modeling, and how it might cause you to modify how you describe the latent variable in your models?

@jim-connolly  You're quite right about Alex's work, Jim.  As soon as I get caught up with the other presentations, I'm going to return to it and engage it more deeply, mostly because I'm likely to learn quite a bit from it.

Your point about how Stowe is read in the 1890's is interesting.  Uncle Tom's Cabin occupies an interesting place in our corpus; it's juvenile reading, but it's read more-or-less equally by boys and girls (see this plot and this one; search for Tom to find the book).  I don't recall that we thought about it all that deeply, since it's neither "girl" nor "boy" reading, other than to cast it as a foil to The Planter's Northern Bride.  If we rework the piece for publication, I'll be sure to raise the issue with the team.  Thanks!

@jim-connolly  Regarding Stowe, Lynne may have more tell us. As I recall, although Uncle Tom's Cabin wasn't prominent in the presentation, early on we started thinking about whether reading Tourgée or Stowe or Chesnutt might represent an ideological bent in opposition to the nascent Lost Cause story. That early guess was overtaken by Lynne's insight that the draw of the romance genre was evidently stronger. Barbara Hochfeld's argument that Stowe was read differently in the 1890s makes a lot of sense. Muncie readers by then were probably also quite familiar with stage adaptations of Uncle Tom's Cabin.

Thanks for the questions, @jfbratt! They're big, so forgive me if my reply has more musings than answers.

Somewhat contrary to my expectation, I found that adding demographics (gender and/or class) did not improve the accuracy or sensitivity of a logistic model of borrower behavior. One cause here has to do with the nature of library policy and use, as Doug, Lynne, and Steve mentioned in their talk: the listed patron gender does not necessarily reflect the person behind the borrowing--indeed, the recorded borrower name might not either. Another cause has to do with the fact that, in a standard logistic model, variables aren't weighted. This means that, say, gender influences the model on the same footing as any other variable (i.e., the number of checkouts for one author).

One can add weights to particular variables in a logistic model (or use some form of dimensionality reduction / principal component analysis on the author checkout variables to reduce their influence relative to demographic variables), but here we would encounter some methodological questions that I'm still undecided on. For one, how confident do we feel in weighting a variable like gender in a dataset where values don't necessarily correspond to the actual borrowing pattern? This is exacerbated by the fact that our best guess at identifying when the recorded gender doesn't reflect the borrower (as Doug, Lynne, and Steve pointed out) is by turning precisely to checkout records for clarification!

I think there are also some larger framing questions that would arise with weighting. If the emphasis is on predicting borrower behavior - as my title, after all, suggests - then any weighting that would improve accuracy and sensitivity across the board would be an improvement to the model. (The risk in setting weights algorithmically for both demographic and checkout variables - which I have not attempted - is that issues of multicollinearity could flare up: namely, it might over-weight checkouts-per-author variables that happen to correlate with demographic variables, which would mean that models would be biased based on how conventionally gendered or classed the activity of reading a particular author was.) But I found myself increasingly interested in the concept of taste in its own right: can we in fact identify consistencies across the rich variability of borrower behavior itself? My assessment, of course, is a qualified yes.

Your question makes me realize that one thing I'd really like to do is compare demographics-based models against behavior-based models as a means of evaluating the degree to which observed taste is or is not independent of observed age, class, or gender.

Genre is a perhaps surprisingly tricky issue to pursue beyond qualitative interpretations on an author-to-author basis, as I did in my presentation. This is partly because turn of the century authors often wrote across genres much more than they tend to today (and patrons followed authors across genres), so not all authors can be associated with one genre. To really shift the focus to genre, what initially seems like the best avenue of approach would be to define checkouts-per-genre as independent variables rather than checkouts-per-author. This would involve going through and tagging each book by genre, which would involve defining subcategories with more nuance than, say, "historical romance"--which would otherwise dominate the data given the genre's ascendance at the time. And that kind of tagging, whatever way it is done, ends up entailing an entire research project in itself.

Thank you Lynne, Steve and Doug, and also Alex for stimulating discussions arising from your employment of the WMR database. I continue to be amazed at the range of uses that are being made of WMR, and continue to hope that there will, in the future, be other such sets of data from other libraries to permit comparative analyses.

If I may make a single point for each of your presentations, I'd like to ask our St. Louis colleagues to consider further the socio-economic and cultural backgrounds of their choice of readers. Both William R. Snyder, who was Muncie's Superintendent of Schools and President of the MPL Board, and Psyche Hoover, the school age daughter of a well-to-do merchant, are classified by us (using Alba Edwards's classification) as High White Collar. My question to you is whether, through your findings, it's possible to show that post Civil War myths of the American South were also shared by readers from lower socio-economic classes?

Alex -- Your discussion of latent variables adds a new dimension to our thinking, and, like Steve, I'd like to listen more closely to your presentation in order to understand its workings. However, I do think it's worthwhile pointing out that the MPL was the designated repository for Delaware County IN of both State and official Federal publications. Most of these were never borrowed, and I am left wondering whether their presence in the library might affect your discussion and statistics of "unborrowed" books.

Thanks @steve_pentecost -- I was really fascinated by your, @douglasknox's, and Lynne's presentation as well!

I don't see borrowers' tendency to check out new books as an issue; it seems to me that new books (or, today, new music or tv) do occupy a large part of taste. And treating taste as a latent variable would seem appropriate to this phenomenon: even though my choices in consuming new novels don't necessarily set or define my taste, they are inherently consistent with my taste (otherwise it wouldn't really be my taste).

The issue of limited holdings (checked-out books) is more of an issue. I've tried to mitigate it by checkouts-per-author as the organization for independent variables rather than checkouts-per-book. I think this is an appropriate workaround because, as several library historians have pointed out, patrons often read by author, which allowed them flexibility when faced with checked-out titles. The likelihood that every held copy of every book by any given author was checked out at any given time is much lower than the likelihood that every held copy of a single book was checked out at any given time. I suggested that title acquisition was responsive to demand and that patrons displayed patience in the long run, but another way of evaluating the size of potential error here would be to measure the average checkout duration for titles in consistent demand.

In logistic regression you can interpret which variables are most associated with a positive outcome for each individual model based on their assigned coefficients. So for example, borrowing books by the plantation romance writers Marion Harland, Mary Jane Holmes, and Augusta Jane Evans are each among the strongest predictors of whether a patron borrowed a book by Caroline Lee Hentz.

One question I wanted to ask your team - and that is related to your question regarding popular new books - is whether you've explored comparative measures for overlap in borrowing. As your presentation noted and as your tables show, many of the most popular books in the late 1890s and early 1900s were historical romances about the Civil War and/or Reconstruction in the South. So we would expect there to be a fair amount of overlap among these books on the bases of genre and of contemporary popularity in addition to subject and setting. Have you compared how patrons who borrowed books in the Southern corpus borrowed other historical romances or bestsellers, for example? I'm wondering about the degree to which an interest in Southern books can be distinguished from an interest in what was currently popular? (I haven't had the chance to read your paper yet, so my apologies if this is already answered there.)

As for the region questions, if I may: Kentucky is somewhat complex in the evolving 19th-century regional system. Timothy Flint very much included it in his original "middle west" (latitudinal middle, longitudinal west) in the 1820s, and Cincinnati's status as a major literary and publishing center in the antebellum period (and, though to a much lesser extent, afterwards) was a primary influence on literary Kentucky. In the 1850s-1870s, the regional formation of "pike country" included Kentucky as well as southern Illinois and Indiana, Missouri, etc.: quite consistent with Flint's categorization. This begins to evolve as Euro-American settlement in the Great Plains and what we'd now call the Upper Midwest (earlier, "the Old Northwest") shifts regional demographic composition, publishing centers, and circulation patterns of cultural and economic materials.

Southern critics did sometimes claim some major Kentucky authors like James Lane Allen or John Fox, Jr., but by several metrics those authors received more attention and sales (relative to population) in the Midwest (and I haven't really seen Southern critics claim certain other leading Kentucky authors in the WMR collection and elsewhere, such as John Uri Lloyd). Similarly, while the important regional magazine the Southern Bivouac published in Kentucky, the state had much stronger ties to networks of Midwestern regional magazines. Some infrastructural links go in the other direction too: the powerhouse popular fiction publisher Bobbs Merrill, based in Indianapolis, had more market share in the South than it did in New England or the Mid-Atlantic. So, I would say that Kentucky and Indiana do have a fairly strong regional literary link at this moment, one that was generally Midwestern but had certain connections to the South as well.

Dunbar not only used both the conventional Black and Midwestern dialects in his poetry--when we look to newspaper discourse and reprinting, it becomes clear that he was also widely recognized as a Midwestern writer, especially in the Midwest. I have an article on this subject forthcoming in American Literary History: what's more, Dunbar mixed the two styles and inspired a wave of Black Midwestern writers and readers. So while there were indeed multiple ways of reading (and not reading) Dunbar, I've no doubt that some of his popularity in Muncie was regionally-driven.

Thanks so much, @azleslie, for weighing in on regionalism with that great response. Your article sounds fascinating.

I'll have some other thoughts on methodological questions in your paper, but don't have all my thoughts in order yet.

To Jim's question about Stowe:  Thank you for the reference to Hochfeld's article. Her findings, as you describe them, sound right to me.  It might be useful to know that taking checkouts over all 8 1/2 years into account, we find that a patron who checked out "Uncle Tom's Cabin" was by far most likely to have borrowed a book in the Elsie Dinsmore series and secondly something by Fosdick from reading we're calling southern reading; from the entire collection, then Finley, Alcott, Alger; Stowe figures largely in juvenile reading patterns, as Steve mentions above.

Alex: As to regionalism, it likely depends on what part of Indiana one is from/is located as to whether Kentucky registers as the same region, but in any case point taken.It might be interesting to take a look at the birth places of the patrons and their parents as an indicator of affiliation or orientation.

And to the question about interest in best sellers--yes certainly some of the readers are tracking best sellers.  Our point is that borrowers (if they actually read the books they checked out) appear to be selecting genre or perceived newness or some other promise of reading pleasure over perceived message, but that the books delivered messages (overtly as in Red Rock or more subtly in the perpetuation of stereotypes that were thereby normalized) nonetheless.