How long is the average dissertation?

Note: Please see the update to this blog!

The best part about writing a dissertation is finding clever ways to procrastinate. The motivation for this blog comes from one of the more creative ways I’ve found to keep myself from writing. I’ve posted about data mining in the past and this post follows up on those ideas using a topic that is relevant to anyone that has ever considered getting, or has successfully completed, their PhD.

I think a major deterrent that keeps people away from graduate school is the requirement to write a dissertation or thesis. One often hears horror stories of the excessive page lengths that are expected. However, most don’t realize that dissertations are filled with lots of white space, e.g., pages are one-sided, lines are double-spaced, and the author can put any material they want in appendices. The actual written portion may only account for less than 50% of the page length. A single chapter may be 30-40 pages in length, whereas the same chapter published in the primary literature may only be 10 or so pages long in a journal. Regardless, students (myself included) tend to fixate on the ‘appropriate’ page length for a dissertation, as if it’s some sort of measure of how much work you’ve done to get your degree. Any professor will tell you that page length is not a good indicator of the quality of your work. Regardless, I feel that some general page length goal should be established prior to writing. This length could be a minimum to ensure you put forth enough effort, or an upper limit to ensure you aren’t too excessive on extraneous details.

It’s debatable as to what, if anything, page length indicates about the quality of one’s work. One could argue that it indicates absolutely nothing. My advisor once told me about a student in Chemistry that produced a dissertation that was less than five pages, and included nothing more than a molecular equation that illustrated the primary findings of the research. I’ve heard of other advisors that strongly discourage students from creating lengthy dissertations. Like any indicator, page length provides information that may or may not be useful. However, I guarantee that almost every graduate student has thought about an appropriate page length on at least one occasion during their education.

The University of Minnesota library system has been maintaining electronic dissertations since 2007 in their Digital Conservancy website. These digital archives represent an excellent opportunity for data mining. I’ve developed a data scraper that gathers information on student dissertations, such as page length, year and month of graduation, major, and primary advisor. Unfortunately, the code will not work unless you are signed in to the University of Minnesota library system. I’ll try my best to explain what the code does so others can use it to gather data on their own. I’ll also provide some figures showing some relevant data about dissertations. Obviously, this sample is not representative of all institutions or time periods, so extrapolation may be unwise. I also won’t be providing any of the raw data, since it isn’t meant to be accessible for those outside of the University system.

I’ll first show the code to get the raw data for each author. The code returns a list with two elements for each author. The first element has the permanent and unique URL for each author’s data and the second element contains a character string with relevant data to be parsed.

#import package

#starting URL to search<-''

#output object

#stopping criteria for search loop
stp.txt<-'2536-2536 of 2536.'

#initiate search loop


	names.tmp<-xpathSApply(html, "//table", xmlValue)[10]
	names.tmp<-gsub("^\\s+", "",strsplit(names.tmp,'\n')[[1]])

	url.txt<-strsplit(names.tmp,', ')


			#get permanent handle
			url.tmp<-gsub(' ','+',x)
			str.tmp<-str.tmp[grep('handle',str.tmp)] #permanent URL

			#parse permanent handle
			perm.tmp<-xpathSApply(perm.tmp, "//td", xmlValue)


	#append data to list, will contain some duplicates

	#reinitiate url search for next iteration<-strsplit(rev(names.tmp)[1],', ')[[1]]<-gsub(' ','+',<-paste(


#remove duplicates

The basic approach is to use functions in the XML package to import and parse raw HTML from the web pages on the Digital Conservancy. This raw HTML is then further parsed using some of the base functions in R, such as grep and strsplit. The tricky part is to find the permanent URL for each student that contains the relevant information. I used the ‘browse by author’ search page as a starting point. Each ‘browse by author’ page contains links to 21 individuals. The code first imports the HTML, finds the permanent URL for each author, reads the HTML for each permanent URL, finds the relevant data for each dissertation, then continues with the next page of 21 authors. The loop stops once all records are imported.

The important part is to identify the format of each URL so the code knows where to look and where to re-initiate each search. For example, each author has a permanent URL that has the basic form plus ‘handle/12345’, where the last five digits are unique to each author (although the number of digits varied). Once the raw HTML is read in for each page of 21 authors, the code has to find text where the word ‘handle’ appears and then save the following digits to the output object. The permanent URL for each student is then accessed and parsed. The important piece of information for each student takes the following form:

University of Minnesota Ph.D. dissertation. July 2012. Major: Business. Advisor: Jane Doe. 1 computer file (PDF); iv, 147 pages, appendices A-B.

This code is found by searching the HTML for words like ‘Major’ or ‘pages’ after parsing the permanent URL by table cells (using the <td></td> tags). This chunk of text is then saved to the output object for additional parsing.

After the online data were obtained, the following code was used to identify page length, major, month of completion, year of completion, and advisor for each character string for each student. It looks messy but it’s designed to identify the data while handling as many exceptions as I was willing to incorporate into the parsing mechanism. It’s really nothing more than repeated calls to grep using appropriate search terms to subset the character string.

#function for parsing text from website

	#separate string by spaces<-strsplit(gsub(',',' ',,fixed=T),' ')[[1]]<-gsub('.','',,fixed=T)

	#get page number
	if(grepl('appendices|appendix|:',pages)) pages<-NA

	#get major, exception for error
		major<-paste(major[nchar(major)>0],collapse=' ')

		}))=='try-error') major<-NA

	#get year of graduation
	if(!length(yr)>0) yr<-NA

	#get month of graduation
	if(!length(month)>0) month<-NA

	#get advisor, exception for error
		advis<-paste(advis,collapse=' ')
		}))=='try-error') advis<-NA

	#output text


#get data using function, ran on 'dat'

#convert to dataframe

#reformat some vectors for analysis

The section of the code that begins with #get data using function takes the online data (stored as dat on my machine) and applies the function to identify the relevant information. The resulting text is converted to a data frame and some minor reworkings are applied to convert some vectors to numeric or factor values. Now the data are analyzed using the check.pgs object.

The data contained 2,536 records for students that completed their dissertations since 2007. The range was incredibly variable (minimum of 21 pages, maximum of 2002), but most dissertations were around 100 to 200 pages.

Interestingly, a lot of students graduated in August just prior to the fall semester. As expected, spikes in defense dates were also observed in December and May at the ends of the fall and spring semesters.

The top four majors with the most dissertations on record were (in descending order) educational policy and administration, electrical engineering, educational psychology, and psychology.

I’ve selected the top fifty majors with the highest number of dissertations and created boxplots to show relative distributions. Not many differences are observed among the majors, although some exceptions are apparent. Economics, mathematics, and biostatistics had the lowest median page lengths, whereas anthropology, history, and political science had the highest median page lengths. This distinction makes sense given the nature of the disciplines.

I’ve also completed a count of number of students per advisor. The maximum number of students that completed their dissertations for a single advisor since 2007 was eight. Anyhow, I’ve satiated my curiosity on this topic so it’s probably best that I actually work on my own dissertation rather than continue blogging. For those interested, the below code was used to create the plots.

#plot summary of data

txt.val<-paste('mean = ',mean.val,'\nmed = ',med.val,'\nsd = ',sd.val,
	'\nmax = ',rang.val[2],'\nmin = ', rang.val[1],sep='')

#histogram for all
hist.dat + geom_histogram(aes(fill=..count..),binwidth=10) +
  scale_fill_gradient("Count", low = "blue", high = "green") +
	xlim(0, 500) + geom_text(aes(x=400,y=100,label=txt.val))

#barplot by month<-ggplot(check.pgs,aes(x=month,fill=..count..))

pdf('C:/Users/Marcus/Desktop/month_bar.pdf',width=10,height=5.5) + geom_bar() + scale_fill_gradient("Count", low = "blue", high = "green")

#histogram by most popular majors
#sort by number of dissertations by major

for(val in 1:length(get.grps)){

	pop.maj<-check.pgs[check.pgs$major %in% pop.maj,]<-aggregate(pop.maj$pages,list(pop.maj$major),function(x) round(median(x)))

	hist.maj<-ggplot(pop.maj, aes(x=pages))
	hist.maj<-hist.maj + geom_histogram(aes(fill = ..count..), binwidth=10)
	hist.maj<-hist.maj + facet_wrap(~major,nrow=2,ncol=2) + xlim(0, 500) +
		scale_fill_gradient("Count", low = "blue", high = "green")

		lab=paste('med =',$x,'\nn =',pop.n$x,sep=' ')

	hist.maj<-hist.maj + geom_text(data=txt.dat, aes(x=x,y=y,label=lab))<-paste('C:/Users/Marcus/Desktop/group_hist',val,'.pdf',sep='')



#boxplots of data for fifty most popular majors

pop.maj<-check.pgs[check.pgs$major %in% pop.maj,]

box.maj<-ggplot(pop.maj, aes(factor(major), pages, fill=pop.maj$major))
box.maj<-box.maj + geom_boxplot(lwd=0.5) + ylim(0,500) + coord_flip()
box.maj + theme(legend.position = "none", axis.title.y=element_blank())

Update: By popular request, I’ve redone the boxplot summary with major sorted by median page length.


Stealing from the internet: Part 1

Well, not stealing but rather some handy tools for data mining… About a year ago I came across the package XML as I was struggling to get some data from various web pages. The purpose of this blog is to describe how this package can be used to quickly gather data from the internet. I’ll first describe how the functions are used and then show how they can be included in a custom function to quickly ‘steal’ what we need.

I realize that data mining from internet sources is not a new concept by any means (a more thorough intro), but the practice is new to me and I have found it very useful in the applications I’ve attempted. For example, I routinely collect supporting data for large datasets that describe lakes in Minnesota. A really useful website that contains a lot of this information is LakeFinder hosted by the Minnesota Department of Natural Resources. The website can be used to access all sorts of information about a lake just by punching in a lake name or unique 8-digit id number. Check out Lake Minnetonka (remember Purple Rain??). The page has lots of information… lake characteristics, fish surveys, consumption warnings, etc. Also note the URL. The last 8 digits are the unique lake ID for Minnetonka assigned by MNDNR. This part of the URL comes into play later.

What if I want information for a couple dozen lakes or even several hundred (yea, or over 10,000)? We have a few options. Obviously we could go through lake by lake and copy the information by hand, but this causes headaches and eventually a desire to harm one’s self. We could also contact the site administrator and batch request the data directly, but this may take some time or may require repeated requests depending on our needs. As you’ll probably soon realize, I hate doing things that are inefficient and the XML package provides us with some useful tools to quickly gather information from the internet.

I’ll walk through an example that shows how we can get maximum depth of the lake by accessing the HTML directly from the website. As a disclaimer, I have very little experience with HTML or any other markup languages (other than LaTeX) so I encourage feedback if the approach below can be implemented more efficiently.

#install and load packages

The XML package has tons of functions and I’m not going to pretend like I understand them all. However, the htmlTreeParse function (or xmlTreeParse) can import raw HTML code from web pages and will be useful for our purposes. Let’s import the HTML code for Lake Minnetonka (remember the last 8 digits of the URL describe the unique lake ID).

html.parse<-xpathApply(html.raw, "//p", xmlValue)

The html.raw object is not immediately useful because it literally contains all of the raw HTML for the entire webpage. We can parse the raw code using the xpathApply function which parses HTML based on the path argument, which in this case specifies parsing of HTML using the paragraph tag.

We now have a list of R objects once we use xpathApply, so we don’t have to mess with HTML/XML anymore. The trick is to parse the text in this list even further to find what we want. If we go back to the Lake Minnetonka page, we can see that ‘Maximum Depth (ft)’ precedes the listed depth for Lake Minnetonka. We can use this text to find the appropriate list element in html.parse that contains the depth data using the grep function from the base package.


It’s not a huge surprise that we can get a return from grep for more than one element that contains ‘depth’. We’ll need to select the correct element that has the depth information we want (in this case, the first element) and further parse the string using the strsplit function.

robj.parse<-robj.parse[[1]] #select first element

depth.parse<-as.numeric(strsplit(strsplit(robj.parse,'ft): ')[[1]][2],'Water')[[1]][1])

The code for depth.parse is really messy but all it does is make two calls to strsplit to grab the depth value based on the text that is directly before and after the info we need 'ft): ' and 'Water', respectively). The final value is converted from a text to numeric object. Seasoned programmers will probably cringe at this code since it will not return the correct value if the web site changes in any way. Yeah, this isn't the most orthodox way of coding but it works for what we need. Undoubtedly there are more robust ways of getting this information but this works just fine for static websites.

Additionally, we can combine all the code from above into a function that parses everything at once.<-function(lake){<-paste(
  html.parse<-xpathApply(html.raw, path="//p", fun=xmlValue)
    strsplit(strsplit(robj.parse,'ft): ')[[1]][2],'Water')[[1]][1]

All we do now is put in the 8-digit lake identifier (as a character string) and out comes the depth. We can make repeated calls to the function to get data for any lake we want, so long as we know the 8-digit identifier. The lake id number is critical because this defines where the raw HTML comes from (i.e., what URL is accessed). Notice that the first part of pastes the input id text with the URL, which is then passed to later functions.

Here's an example getting the information for several lakes using sapply to make repeated calls to


It's easy to see how we can use the function to get data for lots of lakes at the same time, although this example is trivial if you don't care about lakes. However, I think the code is useful if you ever encounter a situation where data are stored online with predictable URLs with predictable text strings surrounding the data you need. I've already used variants of this code to get other data (part 2 on the way). The only modification required to make the function useful for gathering other data is changing the URL and whatever text needs to be parsed to get what you need. So there you have it.

Get all the code here:

#install the packages
#get raw html for Lake Minnetonka
#parse html using <p> tag
html.parse<-xpathApply(html.raw, path="//p", fun=xmlValue)
#find elements that containt 'Depth'
#get correct element with maximum depth info
#use 'strsplit' to get actual value for maximum depth
depth.parse<-as.numeric(strsplit(strsplit(robj.parse,'ft): ')[[1]][2],'Water')[[1]][1])
#combine above code into a single function<-function(lake){<-paste(
html.parse<-xpathApply(html.raw, path="//p", fun=xmlValue)
depth.parse<-as.numeric(strsplit(strsplit(robj.parse,'ft): ')[[1]][2],'Water')[[1]][1])
#examples showing how the function is used'27013300')
view raw gistfile1.r hosted with ❤ by GitHub