| 
  • If you are citizen of an European Union member nation, you may not use this service unless you are at least 16 years old.

  • You already know Dokkio is an AI-powered assistant to organize & manage your digital files & messages. Very soon, Dokkio will support Outlook as well as One Drive. Check it out today!

View
 

Fall09 Class Links, Questions and Answers

Page history last edited by matthewsvogel@gmail.com 14 years, 6 months ago

 

From time to time I will place new class-related links here, and I will make assignments.

In the short run, I suggest that PSA students use this page for questions; and all respondents

should consider providing answers. 

 

Remember, in your readings for this class, try to identify topics you would like to see discussed

in class; also questions you would like to have answered, either here on the wiki, or in class. 

 

For those interested in missing data applications related to PSA, using R, see:

http://www.stefvanbuuren.nl/publications/MICE%20in%20R%20-%20Draft.pdf

 

For a well-written essay on the history and recent forces (circa 2002) affecting educational research see:

http://www.nytimes.com/2002/11/10/education/no-child-left-behind-does-it-work.html?scp=7&sq=traub%20work&st=cse&pagewanted=1

 

Assignment:

In addition to your readings, as discussed on Wednesday, I suggest that each of you prepare the following for

the next class (Tuesday, noon):

1. Go to any source that interests you (Journal or Magazine article, book, newspaper, etc.) to find two or, better,

     three examples where authors make clear inferences of causation, but where the available evidence (is it cited?)

     supports little (or nothing) more than association. That is, find examples where authors make this kind of error.

2.  Explain how selection bias might have been present in at least one of the situations you identify; and for this be

      as specific as you can.

3.  Describe what you would suggest could be done to design a study, a sample, an analysis to reduce if not eliminate

     selection bias in the context your example(s) reference. If you have questions about these, make those explicit.

Added later: If you could send in your responses on these by say 9 pm on Monday that would be most helpful. Thanks.

 

Vogel- Crime Causation

Comments (17)

matthewsvogel@gmail.com said

at 1:52 pm on Sep 14, 2009

Should this first assignment be posted here on the Wiki or emailed to you?

Thanks!

bob pruzek said

at 8:43 pm on Sep 14, 2009

Either place is fine, but the wiki affords others the opportunity to see. So better that! BP

ashley viola said

at 7:54 pm on Sep 17, 2009

In the Holland paper, section 3, why is it "not reall correct" that a measurment is usually thought of as the result of a process that is applied to a unit? Why are 2 response variables needed for a patient receivin either a treatment or control? If they receive one or the other then shouldnt they get one measurment for response variable Y?

bob pruzek said

at 10:25 pm on Sep 17, 2009

Ashley, Assuming you are referencing p. 946, bottom-right, just read the formalities carefully. Holland is saying that to EXPLICATE the MEANING of causality, he needs
two variables, not one. In practice there would be only only one response, but to DEFINE the model, TWO variables are needed. Your question refers to 'receiving either a treatment or control', -- which means NOT BOTH. Potential outcomes are (generally) abstractions, and this is what Holland needs to develop the model fully. You last question is answered by consideration of these formalities. Anyone else want to chime in? Do it! BP

ashley viola said

at 11:09 pm on Sep 17, 2009

I understand better now reading it over a few times - the first time through it was all a little confusing!! I think Holland actually does a good job in helping the reader understand causality starting from the beginning concepts and going through step by step. I now think I understand the meaning behind it! Thanks!!

matthewsvogel@gmail.com said

at 8:01 am on Sep 18, 2009

I think that Ashley has pointed out one of the most important issues that Holland is explicating and one that most people (myself included) often take for granted. If I understand Ashley correctly, she is basically questioning what Holland refers to as the ‘Fundamental Problem of Causal Inference’. As Bob noted, to truly infer causality we need not only a measure of a treatment effect on a particular unit but also a simultaneous measure of the same outcome under the condition of no treatment within the same unit. In practical applications it’s impossible to observe both outcomes , and therefore we have to rely on certain assumptions when making claims about effects of causes. This point really sets the stage for the rest of this paper.

matthewsvogel@gmail.com said

at 8:33 am on Sep 22, 2009

Here is a link to Judea Pearl's website for those of you interested in reading the 2003 TEST publication - the link on dropbox didn't work for me. I meant to put this up earlier in the week, sorry.

http://bayes.cs.ucla.edu/jp_home.html

matthewsvogel@gmail.com said

at 2:33 pm on Sep 23, 2009

Hey fellow PSA-ers... is anyone willing to share their notes from yesterday with me? Shoot me an email mv9674@albany.edu. I'd appreciate any help!

Matt

Jason Bryer said

at 8:01 am on Sep 28, 2009

From the US Consumer Product Safety Commission (http://www.cpsc.gov/CPSCPUB/PREREL/PRHTML99/99175.html) warns about the effects of co-sleeping. From the second sentence they claim that "placing babies to sleep in adult beds puts them at risk of suffocation or strangulation." These warnings are based on data over a 7 year period indicating there were 515 baby deaths linked to adult beds. First, assuming there a legitimate claim to causality here, it seems unlikely that 515 deaths over a 7 year period would even be statistically significant. However, there are several other logical problems with their claim, most egregiously is the claim that these deaths are indeed the result of co-sleeping. There have been a number of peer-reviewed articles published that examined some of these cases and found that in some cases, co-sleeping seemed to have been other factors that could explain the death such as severe parental obesity and alcohol or drug use.

Since randomization is clearly out of the question, PSA could provide some valuable insight into possible causality. More specifically, instead of using a single covariate and attempting to link it to the outcome, using all the available covariates available would provide a better picture of relationship between co-sleeping, other sleeping behaviors (e.g. drinking, drug use, obesity, etc.), and potential child death.

bob pruzek said

at 10:04 am on Sep 28, 2009

This is partly a test: Are you reading this wiki regularly? If yes, you will be better prepared for class tomorrow if you can answer this question:
What is the difference between a 'fine' and a 'course' balancing score? Rosenbaum and Rubin (1983) make this distinction. Be able to discuss. What are the strengths and weaknesses of each type of balancing score?

Weimiao Fan said

at 10:18 pm on Oct 5, 2009

Regarding the research project, I have two questions.

1. In which form that we do this research project? For example, are we suppose to submit a written paper in the end of semester after the presentation? Or the presentation is the final product of the research project?

2. Does this research project focus more on mathematical methods or practices ? For example, in my weighting project, am I suppose to focus on the math formula and math derivation, or more focus on the application of the methods (e.g., the results of simulated data)?

PS: Could you please put the Peter Austin's paper on weighting on the Dropbox.

Thanks very much!

Weimiao

bob pruzek said

at 9:50 pm on Oct 15, 2009

Wiemiao, Since I answered most of this privately, I simply forgot that your q.s were still outstanding vis-a-vis interests of others: So....
1. A project for this course will entail preparation of a paper that I will help you with, by examining one or more drafts, editing, and then ensuring that
that paper is reasonable to distribute to the class. It could happen two ways, the first a being slightly better option: a. prepare the paper, place it on the wiki at least two days before your presentation, and then present to the class using your paper for background and details (possibly with slides). b. prepare slides for a classroom
presentation, present them in class, and then prepare a paper within perhaps a week, again to place this on our psa wiki. Your presentation should, with discussion, be
timed for, say 45 minutes to an hour, recognizing that plenty of time should be allowed for discussion. In the first case in particular, an addendum to your initial paper should be added to the original, pending the results of discussion and followup in class. For the second case, the discussion would of course be incorporated in the paper.
2. The aim should be to examine a topic that deals with applications of some aspect of matching or propensity score analysis. It need not be heavily mathematical, but if mathematical reasoning or methods are centrally involved you should not hold back on such analytic features. Derivations in most cases can be expected to be available in the literature you cite; but if you have something NEW to offer, by all means feel free to present derivations. Improved practice for some aspect of matching, OR details about an application of your choosing, including if you can, a reasonably complete matching or ps analysis, should be your main aim.
....re: ps, all of Peter Austin's papers are already in the dropbox; see names starting with StatisticsInMedicine(.....).pdf
hope this helps. BP

ashley viola said

at 10:41 pm on Oct 15, 2009

Not sure if everyone knows my project topic yet so here is a quick overview. I am interested in matching and would like to show the class a variety of matching techniques including propensity score matching as well as CEM, exact matching, genetic matching and others. My plan is to first show an R program which goes through the exact, genetic, CEM and other matching techniques with an application to the Lalonde dataset. Then I would like to show a SAS program which does propensity score matching. This way the class can get a good understanding of different matching techniques in more than just one programming language.

Ashley

bob pruzek said

at 10:54 pm on Oct 27, 2009

All, Note that I've added a file here, and in the PSArelated dropbox folder called berk.birthwt2.txt. This file contains the birthwts (bwtt) in ounces for all infants. BP

ashley viola said

at 12:17 am on Nov 3, 2009

Hi all - I have recently posted a draft of my project paper on Matching - the goal of the document is for you all to get a general sense of matching and be able to pull out this document when a matching question arrises! Hopefully I covered some of the major topics in matching and if there is something someone would like to see I can do some more research and always add it to the paper. Included in the paper is an overview of a matching program in R which handles multiple matching techiniques. At the end of the paper which is not yet complete I will use the Lalone data and go through a few matching examples using different matching techniques. Any questions just ask.

I have also posted the birthwt assignment that Dr. Pruzek asked us to complete for tomorrow.

ashley viola said

at 12:19 am on Nov 3, 2009

Oh PS one more thing...I also have information on how to do matching in SAS for those that are interested...I will post those articles as well! But they are not included in my paper just so you know.

matthewsvogel@gmail.com said

at 1:48 pm on Nov 23, 2009

Hey everyone, I've uploaded a draft of my course paper that I will be discussing tomorrow. I apologize for the delay on this, and I'm hoping I've left enough time for everyone to review before tomorrow.

You don't have permission to comment on this page.