|Finite mixture models for linked survey and administrative data: Estimation and postestimation
Researchers use finite mixture models to analyze linked survey and
administrative data on labor earnings (or similar variables), taking
account of various types of measurement error in each data source.
... (Read more)
Different combinations of error-ridden or error-free observations characterize latent classes. Latent class probabilities depend on the probabilities of the different types of error. I introduce a set of Stata commands to fit a general class of finite mixture models to fit to linked survey–administrative data. I also provide postestimation commands for assessment of reliability, marginal effects, data simulation, and prediction of hybrid earnings variables that combine information from both data sources.
The Levy Institute
London School of Economics
|Custom estimation tables
In this presentation, I build custom tables from one or more
estimation commands. I demonstrate how to add custom labels for
significant coefficients and how to make targeted style edits to
cells in the table. I conclude with a simple workflow for you to
build your own custom tables from estimation commands.
|Herd effects of topical antibiotic prophylaxis among ICU patients: Simulating a cluster randomized trial using published studies
Importance: Topical antibiotic prophylaxis (TAP) appears to have
potent direct effects towards preventing ventilator associated
pneumonia (VAP) and other infections acquired by intensive care
unit (ICU) patients within randomized concurrent controlled
... (Read more)
Whether TAP engenders indirect (herd) effects on concurrent ICU patients is of great interest.
Objective: To estimate the indirect effects of TAP among concurrent control group patients using data from six broad categories of VAP prevention arranged to simulate a cluster randomized trial (CRT).
Design, setting and participants: Control and intervention groups from 210 published RCCTs from 14 Cochrane reviews of VAP prevention interventions among ICU patients excluding RCCTs with fewer than 50% of patients receiving mechanical ventilation (MV).
Exposures: Indirect exposure of control groups within TAP RCCTs versus within RCCTs of other VAP prevention interventions in five broad categories.
Main outcomes and measures: The proportion of patients with VAP or ICU mortality (primary outcomes), aggregated separately for control and for intervention groups, as abstracted towards deriving study level intervention effect size estimates within each Cochrane review.
Results: Among 14 Cochrane reviews, the strongest summary effect sizes were apparent for VAP and mortality prevention among TAP RCCTs. Surprisingly, the ICU mortality among TAP intervention groups (23.3 [95% CI, 19.5–27.5]) versus an intervention group benchmark derived from the five other categories of intervention (21.1 [95% CI, 19.0–23.3]), respectively, were similar. By contrast, the ICU mortality among TAP control groups was higher versus a control group benchmark derived from the five broad categories of intervention, being 27.9 (95% CI, 23.5–32.5) versus 21.3 (95% CI, 19.2–23.2), respectively. This higher control group mortality remained evident in sensitivity analyses incorporating groups with late mortality, adjusting for group mean age and year of study publication.
Conclusions and Relevance: The control group event rates that underlie the strong summary intervention effect sizes within TAP RCCT’s among patients requiring MV are unusually high. This implicates strong indirect effects from TAP prophylaxis among MV patients and raises concern that the indirect effects to concurrent control group patients, whilst harmful, underlie the appearance of overall benefit.
Ballarat Health Services and The University of Melbourne
|Open panel discussion with Stata developers
|Increasing computational speed by combining Stata and Python
In this presentation, I will discuss ways to increase Stata’s
computational speed by combining it with Python.
... (Read more)
Examples include the comparison of Stata’s ktau command, which requires a calculation time of O(n2) to obtain Kendall’s tau between two variables with sample size n, to my own community-contributed Stata command py_ktau, which implements Python’s algorithm to compute Kendall’s tau with a calculation time of O(nlog(n)). I will also discuss how to use Stata’s profiler and timer commands and provide examples for how to set a seed in Python when running Python code in Stata. Time permitting, I will talk about applications to the search for random permutations.
Australian National University
|Panel discussion: Seasoned users' tips
|Marginal unit interpretation of unconditional quantile regression and recentered influence functions
Unconditional quantile regressions, as introduced by Firpo,
Fortin, and Lemieux (2009), is a special case of recentered
influence functions (RIF) regressions that can be used to relate
how small changes in the distributions of explanatory variables
affect an unconditional distribution statistic of interest.
... (Read more)
While there is general understanding with regards to the analysis and interpretation of changes in continuous variables, difficulties remain when understanding and interpreting dummies that describe qualitative characteristics. On one hand, the implicit interrelationship among binary variables is usually ignored, and on the other hand, standard RIF regressions only capture effects at the margins, not distributional treatment effects. This presentation suggests the use of restricted least-squares regression analysis (Haisken-DeNew and Schmidt, 1997), combined with the use of centered continuous variables and rescaling, to isolate the constant cleanly as the distributional statistic of interest and better interpret the results of RIF regressions in the presence of dummy variables. A Stata ado-file implements this methodology.
John P de New
The University of Melbourne, Melbourne Institute
|Efficient commands for data visualization in large datasets
In this presentation, I discuss the series of custom Stata
commands (PLOT) for efficient visualization of information.
... (Read more)
The PLOT family of commands is particularly useful for visual analyses of admin data, enabling users to produce a variety of highly customizable plots in a fraction of time required by Stata's native graphing commands. Benchmarking of the graphs show that PLOT commands can prove more than 100 times faster than the native commands, with the efficiency gains growing with sample size.
The University of Melbourne
|Using dialog boxes in Stata to collect user parameters for use in a Stata community-contributed command incorporating Python
Stata users often share do- or ado-files making them available for
other users to run the exact same code.
... (Read more)
However, it is often left for the receiving user to update any specific parameters to make the code work for their needs. In this presentation, I show how we created a Stata program that incorporated Python and then built a Stata dialog box around it to allow the end user to be able to quickly and easily update the parameters. This development was used to extract tweets from Twitter and allowed the end user to enter their relevant twitter API keys, the search period, the search term and other relevant data to the search. The code used for this is available on GitHub at https://github.com/SDASANZ/Stata-Python-Twitter-API.
Survey Design and Analysis Services
|Advanced visualizations with Stata II: Complex data structures
This presentation will showcase a new suite of Stata graphs that
can be utilized to visualize complex data structures
(hierarchical, networks, relational).
Vienna University of Economics and Business
A/Prof. Arul Earnest
This half-day workshop is designed specifically for those interested in using multilevel mixed regression models. A brief description of the various analytical approaches to hierarchical data will be shown, including generalized estimating equations and generalized linear latent and mixed models, followed by an application of a linear mixed model to health data using Stata's mixed command. Annotated output from Stata, along with a live demonstration of the software and relevant published journal articles, will be used in the workshop. Participants will be provided a copy of the lecture notes, data, and program files. The format of the workshop is a series of three short lectures followed by quizzes and discussion.