UQRUG 41
Questions: stats, network analysis, NetCDF, spatial, normality of residuals, intra class correlations, plotting multiple linear regression, reverse geocoding, LD Analysis with genepop
2023-08-30: UQRUG 41
R Overview of the Month
This month at UQRUG, Valentina will be going over running some basic statistical analyses using R.
Find more details, and the code here: https://uqrug.netlify.app/posts/2023-08-30-august-stats/
Attendees
Add your name, where you’re from, and why you’re here:
| Name | Where are you from? | What brings you here? |
|---|---|---|
| Nicholas Wiggins | Library | Here to help! |
| Valentina Urrutia Guada | Library | Here to help :) |
| Raul Riesco | ACE | Learning/helping |
| Kar Ng | UQ | paste0(rep(“learning”, 3), “!”) |
| Cameron West | Library | Here to learn! |
| Kim Henville | EAIT | General lurk |
| Marguerite King | SHRS PhD student | Here to learn! |
| Yufan Wang | HABS PhD student | here to learn how to use R for bayesian analysis |
| Abigail O’Hara | HASS PhD student | General knowledge |
| Ee Liz Puah | HABS | Learning |
| Maca | ISSR | Learn |
| Bernadette | library | AHHH |
| Lucas | Science fac | Learn and solve a question |
| Jay | computer science | using RStudio for statistical analysis |
| Jocelyn | UQCCR | Loving the support from this group! |
| Ryan | SOE | Learning |
| Chris Mancini | EAIT Civil Transport Engineering | Also love the support from this group :-) |
| Luke Gaiter | Library | Trying to fix the in room computer & Learning |
Questions
Q1 - What are some good network analysis packages? - Pradeepa/Nick
We’re after some good network analysis packages in R to create a dynamic network of the Australian food system. Already aware of Bootnet, igraph, visNetwork and threejs, but haven’t explored them a lot yet. Looking for more options and opinions on packages.
Answers
- Raul: I think the most used is igraph, I also experimented with networkD3, it has smoother outputs but it has less options. Anyway, all of them are quite difficult to use, for me it was a lot of trial and error to get the simplest of networks >_<. In general I’ll recomend igraph as it covers most of basic needs.
Q2 - What are some good R packages for bayesian analysis? (Prior distributions) - Yufan
I would like to know if there are any online resources that teach you how to fomulate prior distributions using R.
Answers
- https://search.r-project.org/CRAN/refmans/surveil/html/priors.html
- rstanarm - https://mc-stan.org/rstanarm/reference/priors.html
- https://search.r-project.org/CRAN/refmans/BayesCombo/html/plot.PPH.html
- https://cran.r-project.org/web/views/Bayesian.html
Q3 - How to transform a .cvs file into an .nc file - Lucas
I have a .cvs file with contains temperature values per month. Also it has attached lon and lat values. I need those values on QGIS, but i need them separate by month into a raster file.
Answers
- Nick - I’ll follow up with you on this later. We may be able to explore using the stars R package to achieve this
Q4 - Normality of Residuals or Data points? - Kar
To test for one of the assumption - data normality before deciding the suitable omnibus and post-doc test, which “normality” are we looking at? Residual normality or Data value normality
Answers
- Raul: In general, Residual normality is used to test if your data adjust to a linear model. When a test (like ANOVA) has a normality assumption is ussualy refer to normality of data (value). But it will depend on the test.
- Kar: Thank you very much Raul.
Q5 - How do you calculate intra class corelations? - Bernadette
Answers
- Using the irr package which has a icc function. Then you simply feed that the dataframe and the parameters/model you’re using to calculate the corrlelations.
Q6 - For multiple linear regression, how can we plot it and what’s the graph look like? - Jay
Answers
- Raul:Maybe this will help? https://rpubs.com/bensonsyd/385183 “however, in many multiple linear regression situations, the variables we are using cannot be simultaneously represented two-dimensionally so exploring the data visually is far more difficult”. All good. It looks like we can’t really graph multiple linear regression easily. Thanks a lot.
Maybe explore these visualisations? https://statsandr.com/blog/multiple-linear-regression-made-simple/#visualizations-1
- If there are interactions between multiple variables then you can’t visualise that easily.
- You would have to visualise the plot at different levels of the interaction, as the slope and corrleation would be different at every level of the interaction
- If you have 2 independent variables, you need three dimensions to visualise each variable, and their interaction. OR have a 2D plot at each level of the interaction - like slices through each of those
- The only time this is fine is if there is no interaction between the variables, and you can collapse that extra dimension.
Q7 - Does anyone know of good R packages for reverse geocoding? - Chris
For example nearest street address to a GPS coordinate
Answers
Q8 - Does anyone know how to conduct LD analysis using genepop in R? - Ola
Answers
- Raul: Not familiar with genepop package, but according to the manual, they have a LD (linkage desequilibrium) test. The function is
test_LD(
inputFile,
outputFile = "",
settingsFile = "",
dememorization = 10000,
batches = 100,
iterations = 5000,
verbose = interactive()
)