The Strange Powers of the Placebo Effect

The placebo effect is well known but not widely understood. This is a nice summary of how its effectiveness varies in strange ways.

Secrets and lies: is a lot of medical research bunk?

Social scientists & certainly economists tend to look up to medical research, partly because its where the money is, and also because one of its key methods, the randomized control trial, is seen by many as providing a "gold standard" when it comes to measuring treatment effects - though Deaton, Heckman and others have questioned whether RCTs in economics should enjoy this privileged status.
The public generally tend to hold medical research in even higher respect. Medical researchers are good people, passionately if objectively, pushing back the frontiers of knowledge to help make us better.
So how worrying would it be if much medical research was actually wrong? This conclusion has been emerging from the work of a Greek medical researcher, John Ioannidis, and his team. The causes of this problem are various, including publication bias, and are well known but the scale of the problem is probably not. This article may make you distinctly uneasy.
Part of the problem, that key studies have not been replicated and may be wrong, is not peculiar to medicine. It may well plague the social and behavioural sciences too. This article on the subject won't make you feel any better.

Stata resources for treatment effects

There are a large number of resources within Stata for the estimation of treatment effects. Some are part of official Stata and others are user written that can be easily downloaded.

To estimate regression discontinuity models, there is a download rd due to Austin Nichols. Further details at Nichols, Austin. 2007. "Causal Inference with Observational Data." Prepublication draft available at http://pped.org/stata/ciwod.pdf. It is published in the Stata Journal now I think.

To estimate IV models there are several options in Stata.
ivregress is the main program. A download ivreg2 due to Baum, Schaffer & Stillman is very useful - I recommend it. Make sure you get the latest version. Their paper should be used in conjunction with it: http://ideas.repec.org/a/tsj/stataj/v7y2007i4p465-506.html. xtivreg2 is the equivalent program for panel data.

ivtobit and ivprobit do what their names suggest. If using them, you need to satisfy yourself that they are consistent estimators. Caution is appropriate where the instrumented variable is binary. In the latter case biprobit may be better.

cmp (due to David Roodman) allows you to estimate using MLE a wide range of simultaneous models with combinations of linear and non-linear equations provide they satisfy a recursive structure.

treatreg allows the estimation of what Stata calls "treatment effects models". This is something of a misnomer since it only for a very specific model: a linear regression with an endogenous dummy.

condivreg estimates IV models with a single endogenous variable and provides an exact confidence interval for the slope as opposed to the usual asymptotic one. It is particularly useful if weak instruments are a concern.

For estimating Treatment effects using Propensity Score matching there are several downloads including: psmatch2 (Leuven & Sianesi) which does a wide range of matching estimators and nnmatch which does nearest neighbour matching. psbalance allows you to test covariate balance after matching - something that is recomended.

Regression as a matching estimator: Oaxaca-Blinder rides again

Propensity score matching is a well known method of estimating treatments effects with observational data where the treatment is binary and it is assumed there is no selection on unobservables. To recall: one has data on individuals who have been treated. One would like to form a control who are otherwise identical (on average) but who were not treated.
One could match on characteristics but if the dimension of the X's is high that gets very difficult. It turns out that, due to a famous result of Rosenbaum & Rubin, given a key assumption, matching on the probability of being treated (the propensity) is equivalent.
So the norm is to model this probability with say a logit or probit, estimate the predicted probability and form your control group. There are several ways of doing this.
So what if you used a linear probability model instead? Well it turns out that, like speaking prose, you may have been doing this all along without realizing it. In a recent paper P. Kline shows that such a procedure is equivalent to our old friend the Oaxaca-Blinder estimator well known to de-composers. Aside from being easy to do it has several other nice features like being unbiased in finite samples and ensuring exact covariate balance between the two groups in circumstances where it is not guaranteed by other estimators.
For a nice introduction to matching methods see Conniffe et al.

Causal inference in econometrics: structure vs. program evaluation

Those interested in the JEP symposium linked below but would like something a bit more technical should look at Heckman's recent NBER paper:

Building Bridges Between Structural and Program Evaluation Approaches to Evaluating Policy
J J Heckman , NBER 16110, June
This paper compares the structural approach to economic policy analysis with the program evaluation approach. It offers a third way to do policy analysis that combines the best features of both approaches. We illustrate the value of this alternative approach by making the implicit economics of LATE explicit, thereby extending the interpretability and range of policy questions that LATE can answer.

No fancy econometrics needed

The evaluation of labour market treatment effects is one of the most actively researched areas in applied econometrics. Getting good estimates of policy relevant parameters presents some quite technically challenging issues.
But sometimes you really don't need all that. The story below from yesterday's Sunday Business Post gave quite a staggering example of this. A program to get the long term unemployed back to work cost €39m with 46 participants (i.e. almost €900k each) and apparently did not get any of them back to work.
The article adds "Forfas said the state was spending €970 million on a number of jobs and training programmes, but found that there was ‘‘significant information deficit’’, in terms of being able to measure their ‘efficiency and effectiveness’ ".
Indeed. The resources necessary to do a first class evaluation of such programs would be trivial by comparison.

http://www.sbpost.ie/news/failed-work-programme-cost-state-almost-900k-per-person-48270.html

Turn off the TV

Dynamic Treatment Effect Analysis of TV Effects on Child Cognitive Development
Fali Huang & Myoung-jae Lee

We investigate whether TV watching at ages 6-7 and 8-9 affects cognitive development measured by math and reading scores at ages 8-9 using a rich childhood longitudinal sample from NLSY79. Dynamic panel data models are estimated to handle the unobserved child-specific factor, endogeneity of TV watching, and dynamic nature of the causal relation. A special emphasis is put on the last aspect where TV watching affects cognitive development which in turn affects the future TV watching. When this feedback occurs, it is not straightforward to identify and estimate the TV effect. We adopt estimation methods available in the biostatistics literature which can deal with the feedback feature; we also apply the standard econometric panel data IV approaches. Overall, for math score at ages 8-9, we find that watching TV for more than two hours per day during ages 6-9 has a negative total effect
http://d.repec.org/n?u=RePEc:eab:develo:1532&r=cbe