There are 992 entries in the building index of Condit's The Chicago School of Architecture, and I studied each text entry to which they referred. In fact, those 992 entries do not all refer to buildings (Table M1). Moreover, 40 are alternate names for buildings that appear elsewhere in the index (29 buildings have two names in the index, four have three, and one has four).
Table M1. Entries in index, under buildings, from Chicago School of Architecture that do not indicate buildings.
That leaves 311 index entries referring to buildings, but one entry points to three buildings (the Gage Group, see Table M2), so there are 313 buildings referenced by the index. Only 268 of those are in the Chicago area, and I excluded the 45 buildings outside Illinois from further consideration. I included 25 buildings in Chicago suburbs, including 14 in Evanston and a few from Aurora, Chicago Heights, Kenilworth, Maywood, Oak Park, Riverside, and Wilmette. Apart from the Gage Group, several other entries require some explanation (Table M2).
Table M2. Cases where buildings describe in Chicago School of Architecture do not clearly indicate a single building.
I checked every page that building entries in the index referenced. If there was more than one page, I sought the one devoting the most text to the building. There I counted consecutive sentences which included anything about the building, but only up to the fifth sentence. Beyond that, I counted the number of consecutive paragraphs including any mention of the building, rounding to a whole number (so if one full paragraph plus two-thirds of another, I counted two paragraphs). If the building's description reached a full page, or exceeded three paragraphs, I then only recorded the number of pages rounded to the nearest whole number.
To convert all estimates to page numbers, I assume 20 sentences/page and 4 paragraphs/page, yielding a single quantitative measure of the space Condit devoted to a building. A description of one sentence means treatment = 0.05 pages, while three sentences means treatment = 0.15 pages, and two paragraphs treatment = 0.5 pages. At the briefest, buildings whose name appears without any description were assigned treatment = 0.01. To get treatment = 0.05, there had to be a sentence with some description of the building or the architect named. A table in the text summarizes the treatment provided in the main text for all 268 Chicago-area buildings.
Of the 268 Chicago buildings referenced in the index, there are eight I cannot identify. No address is given in the text or figure legends, and I have not found them in any other source. These are listed in Appendix 2. Of the remaining, my focus is the 181 buildings to which Condit devotes some description, and not the 79 he mentions only in passing. I refer to those 181 as the primary buildings covered in Condit's book; the other 79 are secondary (Table M3).
I used two main sources for finding the demolition year: Condit's book and the recent edition of Randall and Randall (1999). The latter was obviously needed to check on buildings still standing in 1964, when The Chicago School of Architecture was published. To follow buildings after 1999, and for a few cases omitted by both Condit and the Randalls, I used Sinkevitch and Petersen (2014) and consulted Wikipedia, Chicago Tribune archives, the Emporis building database, or online archives of the Ryerson library, and otherwise searched widely on Google. A bibliography lists all the major references; I have many notes from Google searches but have not committed them all to the references.
The important caveat to demolitions dates is that Randall and Randall sometimes give a range of dates over which demolition might have occurred. I list all 31 buildings from Condit's book for which demolition date is not known exactly in Appendix 3. In all cases, I have decided whether they were standing when Condit wrote the book in 1964.
Table M3. 268 Chicago-area buildings are referenced in the building index of The Chicago School of Architecture
Except for eight buildings whose address I cannot locate, I have been able to ascertain the current status (2014) and the status when The Chicago School of Architecture was published (1964). In nearly all cases, Condit gives the status in the book, meaning in 1964, and he is very consistent about using present tense to describe buildings still standing and past tense otherwise. In Table M4, I tally cases where my assertion about the 1964 hinges on the use of past tense; perhaps some of these could be questioned, especially the Moses Building, which does not appear in Randall and Randall (1999). The eight buildings whose location is unknown are also noted in Table M4; two were likely standing in 1964, they are described with the present tense, but the other six have unknown status in both 1964 and 2014.
Table M4. Notes about demolition
The basic statistic is fraction surviving \( \theta = { S \over N } \), where S is the number of buildings extant at any specified time and N is the numboer of buildings that were standing some time earlier. The main complication that arose was uncertainty in demolition date in any of those N buildings. One fully explained example serves best to describe how I handled this. Consider as a sample cohort the \( N= \) 55 primary buildings that were completed during the 1890s and that I can identify, and follow forward their survival. Divide those 55 buildings into three mutually-exclusive and exhaustive groups: 22 still standing as of 2014, 26 demolished with known date, and 7 more demolished but with uncertainty.
The worst uncertainty belongs to the Omaha Apartments, since the only information I have about them is that they were not standing in 1964. The Park Gate Hotel was also demolished by 1964, but I found an image of a postcard dated in 1910 to show that it must have stood at least that long. Two other buildings with uncertainty help illustrate the calculations: the Chicago Beach Hotel came down sometime in 1946-1950, and the Isabella building came down after 2003 but by 2013. Thus, as of 2014, I can state with certainty that \( { 33 \over 55 } = 60% \) remain standing. But for any year up to 1964, three buildings must be dropped from the calculation, and for 2003-2013, one must be. Details of how I arrived at survival rates of those 55 buildings at the end of decades should make this clear:
This approach, excluding buildings from some calculations but not others, uses all the information available but does not create an obvious bias. (I made extensive tests with simulated data to check for a bias.)
Logistic regression is a common and familiar way to examine correlations between a survival rate as a dependent variable and a variety of independent variables as predictors. In standard regression, the dependent variable is numeric, and the error (or residuals) around the prediction are assumed to follow a Gaussian distribution. In contrast, logistic regression uses a binomial error in order to handle data where the response is not numeric, but either of two states (i.e. standing or demolished). I omit further explanation because many text books, web sites, and statistical packages describe logistic regression. I estimated parameters of the regression, and credible intervals, using a Bayesian procedure in the program lmerBayes in the statistical package R (see the CTFS R Package).
I was interested in how survival rate after 1964 correlated with Condit's quantitative treatment of a book as one predictor, and a building's construction era as another. The model can be written
\( \theta \sim \ln(T) + Y \)
Where \( \theta \) is survival, T is treatment, and Y is construction year, and the symbol \( \sim \) designates that \( \theta \) is modeled as a function of the two other variables. I use the natural log to normalize \ (T \), which by itself was highly skewed (many buildings have very short treatment while a few have much longer).
The result of the logistic regression, using 178 buildings built prior to 1910 and whose status in 2014 is known, is
\( logit(\theta) = 0.361 \ln(T) + .0.0503 Y - 94.44 \)
where
\( logit(\theta) = \ln { \theta \over { 1 - \theta } } \)
is the transformation used in logistic regression. This formula produces a predicted survival rate given Condit's treatment T and the year of construction. For example, with \( Y=1890 \) and \( T=1 \) (1 page), the predicted survival is \( \theta = 0.652 \), meaning a building from 1890 to which Condit devoted 1 page had a 65% chance of surviving from 1964-2014. By contrast, with \( T=0.01 \), or one sentence in the text, and the same construction date, the prediction is 26% survival.