A new almost-weekly web course Biostatistics for Biomedical Research
will be launched on October 4, 2019. The course is intended both for consumers of research (understand study design, stat methods, stat interpretation) and those doing research (slightly more so for those who work with statisticians). Statisticians may also enjoy the course and entering into the live discussion on YouTube. The main page for the course is here which will contain registration information in early September.
Some topics requested by @f2harrell
followers on Twitter, which will be considered for future sessions, include those below. Please request other topics or make other suggestions about the course by replying to this post. New topics will be opened in datamethods
for each one-hour sessions.
Ideas for Managing Live Questions and Answers
More than 600 registrants so far plan to participate by watching by live streaming on YouTube. YouTube offers a live chat that is also replayed as if it were live if the recording of the video is watched later. Offline and in-depth discussion will take place on a session’s datamethods
topic. I need input on how best to manage Q&A on live chat. A few of my colleagues have volunteered to help moderate the chat, but I’d like some ideas on how to keep it organized and manageable. In an ideal world, the chat could be organized by sections in the BBR notes that are covered in the session, and participants would add their questions in the appropriate section, but the YouTube chat does not support that. I thought about using an Etherpad instead of YouTube chat to force this type of organization, but EtherPads usually choke with more than 50 simultaneous users. Ideally no one would enter duplicate questions or off-topic questions, and would only add a question when the section to which it pertains is still being presented. Please add your thoughts by replying to this topic.
New YouTube Option: Premiering a Video
YouTube now allows an instructor to premiere an already finished video and to interact with participants in the live chat box while the video is playing. The video and its synchronized chat can be replayed at any time by those unable to participate “live”. As with other modes, participants can post discussions and add questions at any time on the datamethods
topic for the session. One disadvantage is the disconnection of discussion that occurred in the chat box from discussion in datamethods
. Would it be better to just distribute the video without premiering it, and ask all participants to put questions and comments in the datamethods
topic?
The “premiere” feature has the advantage of allowing the instructor to create a more polished video while avoiding technical problems while live streaming, and it would not distract the instructor with chat during the streaming. It has the disadvantage that if a participant pauses the video during viewing, the live chat will be ahead of her when she resumes.
Survey of Course Material Structure
Now that several sessions have been done, with a few variations of video structure, I’d like to get a sense of what participants think of various options. I’ve listed the options below with a few advantages and disadvantages. Once this list is complete I will put a poll on Twitter. For all the options, offline Q&A/discussion can be done on datamethods
.
- YouTube Premiere (described) above where a pre-recorded video is live premiered and the instructor actively engages in the YouTube chat box during the premiere (but only a fraction of participants watch the videos live)
- YouTube live streaming with no precording (less polished, more things can go wrong, instructor can get distracted by the chat; only a fraction of participants watch the videos live)
- Ordinary recorded videos with no simultaneous chat, using only datamethods for offline Q&A/discussion
- Add more audio
.mp3
files to the notes to narrate the slides and don’t make video (one sound file per section; easy to produce and maintain; pdf notes can be improved often without changing the sound recording or new sound bites can be added; view notes with a pdf viewer which provides sharper resolution that videos; click on the audio icon to play narration; without video there is no moving cursor to tell the reader exactly what the instructor is discussing)
Suggested Topics
- covariate adjustment in RCTs and writing stat plans for same
- prediction and decision making in medicine
- graphics
- interactions
- power calculation
- RCT Table 1
- most common errors in RCT interpretation
- confounders
- variable selection
- multicollinearity
- data transformation
- inference vs prediction
- simulation based power calculations
- use of AUC/C-statistic to assess predictive utility of models (esp how NOT to interpret them/misuse)
- measurement error and misclassifiction in covariates
- stat methods for studies currently in the news
- optimal evaluation of model fit
- predictive ability, etc for different classes of models
- practical use of ordinal models and specifically, how to logically interpret their output
- practical overview of how to pick and justify priors for Bayesian models, specifically how to allow interactions to be half in and half out of the model
- regularized models and their practical use in medical studies
- hierarchical modeling for provider performance benchmarking / public reporting
- missing data and multiple imputation
The following mnemonics have been defined in the course notes. Clicking on them in the blue boxes in the right margin of the HTML version of the notes will bring up search results in datamethods
. If the only search result you obtain is this overview topic, this means that no one has referenced that mnemonic in a specific discussion. Please add such references so that the course notes will lead to readers finding discussions that are relevant to specific material in notes.
reg-alt reg-purpose reg-advantage reg-nonpar reg-simple reg-simple-gof reg-percentiling reg-mult reg-mult-h0 reg-mult-h0-partial reg-mult-gof reg-xbinary reg-ancova reg-corr reg-anova reg-ia reg-hte reg-ia-cc reg-val