Data analysis is as divese as the experiments people can do. There is no common or general way that could be proposed. It depends on a myriad of things. The best general tipps I can give are:

- be clear about your problem

- understand your data / your variables

- understand the methods you use

- KISS (keep it simple, stupid!)

- use common sense

- be creative (if you think hard about the problem, you will have own ideas that may be superior to things you find in literature)

- be critical (does the plot really show what it is supposed to, is it understandable, is all and only all relevant information shown?)

Readings:

https://clauswilke.com/dataviz/

https://www.youtube.com/watch?v=fSgEeI2Xpdc

https://g30seminar.files.wordpress.com/2013/07/graphical-excellenceedward-tufte.pdf

https://bookdown.org/paul/applied-data-visualization/

In my opinion this can constitute any set of summary statistics that adequately describes the key features of the data, you can use mean, median and mode as a measure of central tendency and standard deviation, Interquartile range etc as measure of data spread, for both univariate and multivariate data.

An useful measure of shape of data distribution can be obtained through histogram plots, measure of asymmetry in distribution through skewness and kurtosis calculations etc.

For multivariate data, statistics that measure correlation among a pair of variables or a set of variables can be used, PCA can give you total variation in dataset and its partitioning in variation components due to PCs. Linear Discriminant analysis is another such option that allows for data dimensionality reduction and classification.

Self Organizing maps, Cluster analysis and other machine learning tools can also give you useful insight in data visualization/presentation.

Data analysis is as divese as the experiments people can do. There is no common or general way that could be proposed. It depends on a myriad of things. The best general tipps I can give are:

- be clear about your problem

- understand your data / your variables

- understand the methods you use

- KISS (keep it simple, stupid!)

- use common sense

- be creative (if you think hard about the problem, you will have own ideas that may be superior to things you find in literature)

- be critical (does the plot really show what it is supposed to, is it understandable, is all and only all relevant information shown?)

Readings:

https://clauswilke.com/dataviz/

https://www.youtube.com/watch?v=fSgEeI2Xpdc

https://g30seminar.files.wordpress.com/2013/07/graphical-excellenceedward-tufte.pdf

https://bookdown.org/paul/applied-data-visualization/

You can find many data visualization tools in Microsoft Power BI (Business intelligence) software.

https://powerbi.microsoft.com/en-us/

Enjoy with Power BI.

Regards,

Suranga

You can use python, PowerBI or Google dát studio. Depend on ur taste and how ur dataset looks like

For those who are looking for concrete examples rather than general advices, I found that Excel offers very good tools for preliminary analysis. Highlight your data and press the ideas button on the Home tab. It offers many insights and charts that you might want to use including the relationship between variables, histograms. Also, the data analysis option in Excel is very good for initial statistical analysis and tests (including Anova, T-test, Covariance, Correlation, single linear regression...etc.) which I normally implement in Python or R.

Thank you Debopam Ghosh for mentioning LDA which might be also helpful.

Greetings,

To avoid overlap in many of the good responses I will offer a suggestion that is a lesser-used method, Parallel Coordinates Plots. A PCP is an interesting way to visualize data along many dimensions. MATLAB has great built-in functionality for these (https://www.mathworks.com/help/stats/parallelcoords.html). You can also use the classification learner GUI (in MATLAB) to acquire projection clustering plots along with PCPs. You can represent the information within multiple projection clustering plots with a single PCP, as an example I include a figure from a previous work. Here, you can see (a) that projecting along the PCA basis vectors does not provide class separation (at least in the problem posed, PCA does have a lot of utility elsewhere), this also shows that the variance explained in the direction of the top PCA modes (associated with largest eigenvalues) is being packed in the first few PC basis vectors (as its intended to do). With our in-house method (c) you can see class separation along the 4th and 11th basis vectors in the 13 dimensional subspace. So you can tell a large story from a single PCP. In general, I would agree with Jochen Wilhelm as the type of data visualization you choose is solely dependent on the objective of the hypothesis and the story you wish to portray.

Best,

Tyler J Grear

I see a problem with your question

People are always trying to fit in a box: if you remove your unique take on research or your data analysis, you will end up producing another paper that is boring and provides little to no contribution to the community

• Use tools and packages that you like
• Add details that you think make your contribution interesting
• ask people outside of your group if your plots are easy to understand (sometimes plots add an extra layer of confusion)

Debopam Ghosh I see a problem with throwing means and standard deviations like that!

Eiman Kanjo you got to know your data: plot histograms and densities. Understand if a central tendency makes sense to summarize your data, maybe a variable has two modes, so the average makes little sense.

Also the standard deviation may not be informative, but you have to know your data before you can decide how to summarize it