DATE OF LAST UPDATE: 5/12/06
ADVANCED CRIMINAL JUSTICE STATISTICS:
MULTILEVEL MODELS IN CRIMINAL JUSTICE RESEARCH
Main Course Page: http://www.rbtaylor.net/605_sp06_main.htm
Instructor Home Page: http://www.rbtaylor.net
POSTER SESSION ON 4/27 CLICK HERE
PUTTING PAPERS TOGETHER CLICK HERE
BACKGROUND FEAR REFERENCES CLICK HERE
Class Bulletin Board
Important information will be posted here
5/12 - final grades for course and for paper grades are available - go to memos
5/6/06 - Grades for the second in class exam and the actual in class exam have been posted - go to memos
4/26/06 - Memo about second in-class exam (what to expect, how to study), and detailed grading rubric for final paper CLICK HERE
4/18/06 - Additional article for class Thursday CLICK HERE
4/6/06 - go to memos for handout on common points needing improvement in draft methods sections or CLICK HERE
3/28/06 - go to memos for handout on R squared in HLM or CLICK HERE
3/24 See the link above about putting papers together. Please download it, read it, reread it. Let me know if there are comments or questions.
3/22/06 - Handout from last week CLICK HERE or go to memos.
NEW HANDOUT also posted for 3/23 class CLICK HERE or go to memos
NEW MDM FILE POSTED to go with 3/23 handout - CLICK HERE or go to memos
3/2/06 - mini midterm results posted - go to memos or CLICK HERE
2/28/06 - the handouts we used in the last two classes have been posted - they are on the HANDOUTS page.
2/22/06 - memo about the midterm posted
2/16/06 - links to district mdm and data files - go to data page
|Instructor||R. B. Taylor (GH 536-7)|
|Time and Place||THURSDAY 3:00 - 5:30 (+/-)|
|Office Hours||Monday afternoon 1 - 3 and by appt. as needed|
||TEL: 215.204.7169 (v). You also can ring 1-7918 and ask Ms. Salerno
if we need to chat and the phone is not being picked up.
Students who may require special services should notify the instructor at the earliest opportunity, and I will put you into contact with the Office of Disability Resources and Services at Temple (http://www.temple.edu/disability - 215.204.1280). You may require special services if you are sight or hearing impaired, or if you wish to register for gaining extra time for taking exams.
Hierarchical Linear Models (HLM) or Multilevel models (MLM) represent a significant advance in social scientists' ability to understand how outcomes are affected by context, how individual and contextual factors interact, how outcomes change over time, and how to summarize results from a series of studies. We will be concentrating in this course largely on the first two uses of HLM. We will get to the other two uses of HLM - to investigate changes over time, and to summarize studies -- if and as time permits. I am optimistic. These models address a range of theoretical and methodological issues relevant to criminal justice, sociology, psychology, urban studies, education, and political science. The issues include multilevel analysis, aggregation issues, contextual analysis, and clustered samples. In simple, whenever the individual units of study (e.g., students) are nested within a higher level unit (e.g., schools), HLM is an appropriate, and some would argue the most appropriate form of analysis.
HLM refers both to an analytical technique, and a specific software program. There are other multilevel software programs available, like MLWin. In this course we are using HLM.
There are questions about whether MLMs are just a "fad" right now in criminal justice and criminology research. I don't think so, for the following reason. First, many problems in research and evaluation in criminal justice are most amenable to, and only amenable to, a multilevel approach. Offender or delinquency careers represent cases in point. In addition, the interaction between the lower level unit and the higher level unit (e.g., the officer and the police department) is fundamental to numerous theoretical and policy concerns, and MLMs provide a systematic way to approach these.
Here are some examples of "units nested within
larger units" in criminal justice evaluation or research:
|Level 1 units||Level 2 units|
|Police Officers||Different Precincts|
|Police Precincts||Different Police Departments|
|Cases Sentenced||Different Judges|
|Sentenced Drug Offenders||Different Drug Courts|
|Juveniles||Different Treatment Programs|
You will note with the last example that time is nested. This is a repeated observation setup. You will be learning how observations can be nested within the same individual, or the same unit. This means that MLMs can analyze much of the same data analyzed by repeated measures ANOVA, or even time series, and in some cases, depending on the circumstances, do a better job of it.
At the same time, I do NOT think MLMs are going to "solve" all or nearly all of our analytic problems. That would be asking too much. In fact I think many may become disenchanted with MLMs because the answers they provide may not be to our liking. We have many theory and policy ideas around the interaction between person and context. But MLMs may often find that these interactions are nonexistent or trivial in the pragmatic sense. Alternatively, I think MLMs can help "push" us in our theorizing, moving us to think in more detailed ways about processes. For a great example see: Wilcox, P., K.C. Land, and S.A. Hunt. 2003. Criminal circumstance: A Dynamic multicontextual criminal opportunity theory. New York: Aldine deGruyter.
In the long run, my guess is that MLMs will become like SEMs and other general purpose but also somewhat specialized multivariate techniques: very useful in a wide range of situations, but also easily mis-applied
In this class we use examples from criminal justice and other disciplines.
Goals and activities
The goals of this course revolve around different sets of activities. These activities break into two clusters.
The first is concerned with general, doctoral student skills: completing secondary analysis, and writing research papers. You will be asked, in the beginning of this course, to work with the an extract of the Fall 2003 Philadelphia Area Survey using just cases from Philadelphia, about 350. Go to the data page to learn more. You have a file about individual survey respondents, and a file about the characteristics of the neighborhoods in which they work, including crime rates.
You will check these files, and learn about them. These are exactly the steps you take whenever you begin a secondary data analysis project.
With these data, I am particularly interested in concentrating on respondents' attitudes toward the police, and confidence in the criminal justice system. So our key outcomes here are attitudinal variables.
In order to write your research paper, you will be learning some about research in this area. I will provide you with some starter references in this area. For you to complete your papers you will need to complete additional research on your own and find additional references to be incorporated into your work. The references provided are just starters. If you do not know how to conduct Social Science Citation Index Analyses to "grow" your reference list, then let me know, and we will schedule a session for folks.
Because we have a large number of students and a relatively small number of outcomes, there is going to be some jockeying around in terms of different people specifying different conceptual models. In light thereof I also am going to allow you to work with no more than two other classmates as co-authors. If and as a research team grows, so too do my expectations of how good their product will be.
You will be completing portions of the research paper as the semester progresses, and, in a poster session at the end of the semester, you will be presenting your results at an "in house" event. I am happy to continue to work with students past the end of the semester if and as you seek to prepare your work for submission to a conference and, thereafter, even better, for a journal submission.
The second cluster is specific to course content: carrying out and interpreting contextual regressions; understanding the fundamentals of HLM; and carrying out and interpreting HLMs. In addition, we will be reading a small number of articles using HLM, so that you become familiar with interpreting HLM tabular output.
There is a new version - 6 - of HLM. It has significantly expanded capacities and ease of use in some areas. Whether we will get time to go over all this depends.
We will be having some lab time. Not sure at this time exactly how much. There is a student version of HLM you can acquire. Encourage you to do so because then you can do some simple things at home. Details below.
As mentioned above, the area within which you will be working is attitudes toward the police and confidence in the criminal justice system. On the CD you will find a research proposal providing some background on what, IMHO, are some important issues in this area, as well as related readings.
I assume you understand the basics of OLS multiple regression AND RUNNING SPSS, including:
adjusted R squared
F test of R squared
standard errors of b weights
t tests of b weights
tests for linear vs. curvilinear impacts
coefficient of alienation
SAVING DATA FILES
WORKING IN SYNTAX BOXES
BASIC DATA PROCESSING STATEMENTS - computing new variables, transforming variables, and the like.
In addition, I assume you know your way around SPSS for Windows. This includes being able to write syntax boxes, and save them, and diagnose what is happening with them.
If any of the above terms is unfamiliar to you, you have some remedial work to do!.
Readings and books
You will be reading books, articles, and handouts. The two required books have been ordered. Of course, although I officially recommend the College Bookstore for all your collegiate purchases, you also may find you can save money either through Amazon.com or through ssicentral.com's bookstore.
In addition to the books indicated below, there are handout note files prepared by the instructor (see handouts link), and additional works completed in the area.
You also will be reading many articles in your interest
Stephen W. Raudenbush, Anthony S. Bryk (2002). Hierarchical Linear Models : Applications and Data Analysis Methods. SECOND EDITION Thousand Oaks: Sage. This book costs a lot but is is a crucial reference. It is close to $100. Feel free to look for used copies.
Stephen Raudenbush, Anthony Bryk, Yuk Fai Cheong, Richard Congdon, and Mathilda duToit (2004). HLM 6: Hierarchical Linear and Nonlinear Modeling . Chicago: SSI Scientific Software International. This is the program manual. Be sure you get version 6
J. J. Hox (1995) Applied Multilevel Analysis. Available on the web at http://www.rbtaylor.net/hlm_applied_hox.pdf [This is over a hundred pages. You want to save this to disk. I may be making reference to this material from week to week. I like the discussion about connecting across levels. There is some useful, basic material here. Again, you will need to be careful because his Greek differs from the other Greek you will see in other places. Please do not print this all out - just print out pages as you need them.]
This first book says it provides a basic overview. The running example is from political science. I am not sure yet what I think about this book. In the first few pages I have found stuff that is potentially confusing. But try it -- it is cheap at least -- and see if it works for you:
Luke, Dougla A. (2004). Multilevel modeling. (Series: Quantitative Applications in the Social Sciences #143). Thousand Oaks: Sage.
These next two are recommended in that they provide additional detail; they are not necessarily oriented to those who are looking for more basic help.
Ita G. G. Kreft, Jan De Leeuw (1998). Introducing Multilevel Modeling (Introducing Statistical Methods) Thousand Oaks: Sage. [Chapter 2 is a nice review of contextual models. There are some nice graphical examples of varying slopes. The text uses a different program than we are using, so the programming examples are not that helpful.
Tom A. B. Snijders and Roel J. Bosker (1999). Multilevel analysis: An Introduction to basic and advanced multilevel modeling Thousand Oaks: Sage. [More "advanced" [confusing?] than the above book. Good discussion of sampling issues, however. Uses Greek alphabet differently than HLM program does. This book, however, has some stuff you cannot find anywhere else.
ARTICLES AND NOTE FILES
Articles. We are going to be reading a good number of articles
and handouts. These will illuminate uses of HLM, or related and important
conceptual questions. I am going to try and put most of these on a CD,
although some may have to go on a website.
Handouts. The handouts are a series of HLM lecture notes you get directly off the website.
MLM is growing like topsy. Luke (2004) recommends a couple of websites as being among the best. These include
The folks at University of Bristol (UK) maintain the following:
http://www.mlwin.com/ [This used do be: http://multilevel.ioe.ac.uk/]
The folks at UCLA, of whom deLeeuw is one, maintain:
These may or may not prove helpful.
You will complete the assigned readings on a weekly basis and come to class prepared. To help you prepare there are questions to go along with the article readings. You want to write answers to some of those questions after you have read the articles. Some weeks I will announce you can receive extra credit by turning in written answers to questions - but you will not get notice beforehand. Your answers should be typed, double-spaced, with both your name and SSN and due date on the top of the page.
You will complete and write up a number of analytical assignments; the purpose of each is to conduct a specific analysis, and write it up. You will receive full credit for turning in each of these on time provided you make a serious, good faith attempt to complete the assignment. In addition, some assignments ask you to turn in a draft of a portion of your final paper. Each of these assignments will build toward a broader, final paper. These papers will, hopefully, represent building blocks that can later be put together to form a paper of potentially publishable quality. In class I will tell you why it is important to produce something like this.
During the second of third week of class we will be talking about the range of conceptual models that might be applied to these data. This is the first stage in paper project selection. There may be some latitude for students going outside the provided data file and working on a different data source. but you will need to talk to me beforehand.
The weekly assignments are to be typed, double-spaced with the due date your name and SSN at the top. These are not going to be graded, but you will get feedback on how to improve. Be sure to spell-check and grammar-check your papers.
The weekly readings provide the needed conceptual background for carrying out the work assigned, and for understanding how MLMs are advancing scholarship in criminal justice and criminology. Thus it is important that you keep up with the readings. They provide not only examples that help us decode MLMs, they also provide very interesting background for the conceptual work.
You will notify me beforehand if it is absolutely essential for you to miss a class. Given the amount of ground we must cover, a missed class may create a significant burden for your learning curve. If you do miss a class it is completely your responsibility to get all handouts, assignments, and so on, that were distributed.
Your grade at the end of the semester will be based on
|30%||Final paper. It is intended that this be of close-to-submittable quality|
|25%||Final examination. This will focus on the identification of an appropriate tool to use in a particular situation; and on interpreting results presented in tables|
|20%||CREDIT for turning in written assignments on time. These will probably build to the final paper. IF you make a credible attempt to complete the assignment, you get full credit.|
|15%||Short, in-class, mid-term examination|
|10%||Poster session presentation|
Extra credit - see above
1. Assignments are due on the date indicated. If you cannot get your paper in to me at class time, please send me an email explaining why, and let's be sure to have a followup chat. The weekly assignments must not only be credible but also handed in ON TIME in order for you to get full credit. Same applies to the final paper.
2. If I encounter solid evidence of academic misconduct I reserve the right to fail you on the assignment in question, and/or to assign you a failing grade for the course. I will try to state as clearly as I can the ways in which it is acceptable for you to cooperate with one another and network, and the ways in which it is not acceptable.
3. You do have a right to submit any assignment for regrading. (In this course, this means the midterm, final, and paper.) You should state in writing the reason you think you deserve a higher grade, attach that to the original completed assignment, and return it to me. Your grade may go up, go down, or stay the same. I may consult with other faculty members as I deem fit.
GUIDELINES ON AVOIDING ACADEMIC MISCONDUCT
HERE to see College Policy circa 1983 - I think this gives you the most
detail. STRONGLY RECOMMENDED.
We will discuss in class the nature of academic misconduct, including plagiarism. You are responsible for understanding the different varieties of academic misconduct, and for understanding the Graduate School's policies as described below. If I encounter solid evidence of academic misconduct I will discuss the matter with you, and then deliver the consequence I deem appropriate. Possible consequences include: failure on the assignment in question (i.e., a 0); assigning a failing grade for the course; or attempting to have you expelled from Temple University. Should you wish to contest a decision I make on academic misconduct, I will inform you of the procedures to follow. The department and the college have fully specified grievance procedures for graduate students.
The following materials are from the University's Graduate Bulletin statements on academic honesty of about four years ago [http://isc.temple.edu/grad/Bulletin/Default.htm] Even though I can no longer find this in the current Graduate School Bulletin, it still applies.
Temple University believes strongly in academic honesty and integrity; therefore, any kind of academic dishonesty is prohibited. Essential to intellectual growth is the development of independent thought and of a respect for the thoughts of others. The prohibition against academic dishonesty is intended to foster this independence and respect. Primarily, the two types of academic dishonesty include the following: Plagiarism and Academic Cheating.
Plagiarism is the unacknowledged use of another personís labor, ideas, words, or assistance. Normally, all work done for courses ó papers, examinations, homework exercises, laboratory reports, oral presentations ó is expected to be the individual effort of the student presenting the work. There are many forms of plagiarism: repeating another personís sentence as your own, adopting a particularly apt phrase as your own, paraphrasing someone elseís argument as your own, or even presenting someone elseís line of thinking in the development of a thesis as though it were your own. All these forms of plagiarism are prohibited both by the traditional principles of academic honesty and by the regulations of Temple University. Our education and our research encourage us to explore and use the ideas of others, and as writers we will frequently want to use the ideas and even the words of others. It is perfectly acceptable to do so; but we must never submit someone elseís work as if it were our own, rather we must give appropriate credit to the originator.
Academic Cheating is, generally, the thwarting or breaking of the general rules of academic work or the specific rules of the individual courses. Some examples include: falsifying data; submitting, without the instructorís approval, work in one course that was done for another; helping others to plagiarize; or cheating from oneís own or anotherís work; or actually doing the work of another person.
The penalty for academic dishonesty can vary from a reprimand and receiving a failing grade for a particular assignment, to a failing grade in a course, suspension, or expulsion from the University. The penalty varies with the nature of the offense, the individual instructor, the department, and the school or college.
For more information about what constitutes Academic Dishonesty or about disciplinary and/or academic grievance procedures refer to the Universityís Statement on Academic Honesty and the Student Code of Conduct or contact the Student Assistance Center, 215-204-8531.
This course ends up being somewhat more demanding than some other graduate courses for some students. In short, for some of you, this may "feel" like a four credit or a six credit graduate course. Try to plan your weeks (and weekends) accordingly.
For HLM we will use a specific program, put out by Scientific Software International. The main web page for SSI is SSICENTRAL.COM. This is a useful website, because you can look at the examples, and get help interpreting HLM output. I STRONGLY ENCOURAGE YOU TO GO THROUGH ALL THESE EXAMPLES ONCE WE ACTUALLY START WORKING ON HLM. Here is the link:
The department has bought a site license for this program. This means that you need to do your work in the department on their computers if you want to have access to the FULL version of the program
There is, however, a STUDENT version of the file. See the SSCICENTRAL website for more details. I will be putting a version of the student program on your CD. If you want to go to SSICENTRAL in the meantime and download it yourself, you can do that too.
There are, however, a number of restrictions with the student model. Most importantly, you can not run very complex models Below are quotes from a paper I can no longer find that described the limitations of the student edition.
The student edition can run all the analyses the full version can in terms of models selected, statistical options and output.
Restrictions are, however, placed on the data used and the size of the model selected. The following restrictions apply in this edition:
The student edition can run all the analyses the full version can in terms of models selected, statistical options and output. Restrictions are, however, placed on the data used and the size of the model selected. The following restrictions apply in this edition (ACCORDING TO THE WEBSITE):
When these limitations are exceeded, an appropriate error message will automatically be displayed.