Statistical Computing Seminars
Statistical Writing
- Research is a process
- The better the quality of the earlier stages, the better the quality
of the results section
- General discussion of what needs to happen before you start writing
- Careful planning of each stage of the research
- Power analysis (or a detailed analysis plan)
- The goal is to reduce panic
- Use grad school as an opportunity to learn new statistical
techniques and software packages
- How long ago was your last stats class?
- Real data can be real messy
- Four issues that are often problematic - 1. Missing data
- Learn pros and cons of various techniques
- No hard-and-fast rules
- Package to do the imputation
- Package to analyze the imputed data set(s)
- Four issues that are often problematic - 2. Small sample sizes
- Common stat techniques are often not appropriate
- Model may not run for numerical reasons
- Assumptions may not be met
- "Fall back" to a simpler technique
- Issues of fair and accurate reporting
- Four issues that are often problematic - 3. Survey data
- Not like data from experiments
- Different commands or a different stat package
- Four issues that are often problematic - 4. Correlated data
- Patients or doctors (nested) in hospitals, etc.
- Several possible ways to analyze correlated data
- May have to analyze the data in multiple ways
- The results section is an extension of the methods section
- Clear and precise analysis plan
- Changing analyses
- Need to understand the stat techniques and their assumptions just as
you need to understand your substantive area
- A lot of work between getting the appropriate output and being ready
to explain the results to others (AKA writing the results section)
- ANOVA example
- Specifics
- Set up the research question clearly and precisely
- Establish the reader's expectations
- What to leave in and what to leave out? Tell a story
- No relationship between the amount of time something took and how
much space in the write-up it gets
- Careful balance between enough detail to replicate the experiment
and space limitations imposed by the journal
- Describing your data
- Avoid including p-values in the description of the data
- Purpose: description, not hypothesis testing
- alpha inflation
- With 10 tests, the nominal alpha level is .40, not .05
- Planning, fair and accurate reporting
- Reproducibility
- Analyses
- Order your hypotheses (and hence, analyses) from most to least
important
- Confusion about the meaning of a specific p-value
- False precision
- Statistical significance versus clinical relevance
- Report effect sizes
- Where to find examples
- Articles that report similar analyses
- Write-ups can vary widely between disciplines
- DAE pages, annotated outputs, Long and Freese (2006)
- Examples
- Coding dichotomous variables - meaning of coefficient and interaction
term
- Categorical predictor variables
- Logistic regression
- Confidence intervals
- Interaction terms
- Bivariate tests
- Watch words
- chance
- odds
- risk
- probability
- significance (statistical or clinical, parameter or model)
- likelihood
- standardized (variable, coefficient, test scores)
- normal
- controlling for (this is an idea that is in the analyst's head, not the
program analyzing the data)
- covariates
- robust (regression, standard errors, findings)
- nested, hierarchical (models, data)
- random (variables, intercepts, slopes, effects)
- Tables and graphs
- Can be more difficult than it seems to create clear and useful tables
and graphs
- Clarity is paramount; less is more
- Several useful references are available
- Distinction between what goes in the results section and what goes in
the discussion section (discipline specific)
- Things to avoid
- "more significant" (use effect size or omega-squared instead)
- "almost significant"
- Don't bother speculating about why non-significant results are
non-significant
- Just say no to post-hoc power analyses
- Future trends - Increasing sophistication of statistical analyses
- In times past, theory and/or software was not available
- A "fancier" model is not necessarily better
- Good match between the data and the analysis technique
- Missing data example
- Future trends - The use of the web
- Researchers or journals posting data and syntax on web site
- Use syntax (instead of point-and-click)
- Useful if you need to replicate or modify an analysis for an R&R
- Document data transformations, analyses and thought process
- Honesty about the number of significance tests run on the data
- Resources mentioned in this presentation
- Online Seminars including
- Power Analysis
- Statistics Books for Loan
- Applied Statistics Courses
Offered at UCLA
- Statistical Consulting Schedule
- Statistical Consulting Services
The main page for this seminar is here .
UCLA Researchers are invited to our Statistical Consulting Services
We recommend others to our list of Other Resources for Statistical Computing Help
These pages are Copyrighted (c) by UCLA Academic Technology Services
The content of this web site should not be
construed as an endorsement of any particular web site, book, or software
product by the University of California
