How do you generate and test hypotheses with visualization?

I get what this is saying, but it is about more than just understanding the data and the technical context. I think you need to understand the purpose and the business context. Rarely are we told, "Here is all the company data; come up with a hypothesis." There is a purpose for actions. It may be as generic as "Improve Supply Chain" or specific on "Improving our ability to do a revenue forecast," but there is a purpose. This is important to understand since this is your scope. The business context is equally important since data, with context, can become information. Nothing else listed here matters unless you understand how it applies to your organization.

I would say, from business lens perspective, understanding the data through visualization is not merely an analytical exercise but I would say a strategic imperative. It's a foundational step for generating hypotheses that could drive growth, enhance customer experiences, reduce risks, and improve overall business performance. While testing the hypotheses, visualization could serve as a form of evidence to support or refute underlying claims. One may use data visualizations to illustrate patterns, differences, or trends that support the given hypotheses. One may create visualizations to check if the data is normally distributed, independent, or homoscedastic. And, in case assumptions are violated, it may lead to revised hypotheses.

Hmm ... this is not how I research or approach a problem. I usually start with brainstorming and interviewing. I don't consider these as really hypotheses yet, since they are (usually) what people THINK, and may have no evidence or bearing on what can be observed in the data. I consider these assumptions which give me more of the business context and (maybe) avenues to investigate. In any case, need the assumptions, however, to know what data MAY be needed. I then look at the data itself and generate different hypotheses based on observations. In my experience, what people assume is happening is NOT what the data shows. This is valuable all by itself since it opens up people to think out of the box.

An example of testing hypothesis in cancer and environmental epidemiology is Kaplan-Meier survival graphs with log rank tests. It can also be supplemented with Dose-response relationship graphs including confidence intervals (its similar to box and whiskers plot and can be found in excel under "Stock" graphs). The graphs and plots will help understand if you fail to reject alternative hypothesis. I've utilized different scenarios utilizing same dataset i.e. 1,3, and 5 years survival for 1, 3, and 5 years before diagnosis exposure. The graphs were additionally plotted for both single and multipollutant models adjusting for model covariates, nearer monitor distance with fewer missing exposure values to test the assumption and confounding.

I would say, testing of hypotheses is often an iterative process which involves going back & forth between visualization, statistical analysis, and hypothesis refinement. In most cases, the visualization plays a central role in the iterative process, helps to explore the data and validate the given hypotheses thoroughly. E.g., In manufacturing, where process operating within control limits or not, a control chart which tracks a specific quality metric over time could be very useful. Assuming data points fall within control limits and show no unusual patterns, it supports the hypothesis that the process is in control. Likewise, ANOVA plot can be helpful in validating the multiple groups have significantly different means.

Testing your hypotheses, especially if you are working in a team environment, may be valuable to the business overall. One of my teams worked on an analysis of EHS data and found that the assumptions and initial hypotheses were not represented in the data. This testing led to new approaches, data collection, and visualizations still being used today.

Visualization helps in refining the hypotheses by confirming, challenging, or expanding the initial ideas based on the evidence within the data. During iteration visualization could reveal contradictory or surprising findings which may question the initial hypotheses as new data patterns might have come out. This turns into opportunity to refine your hypotheses. During exploration do notice the sensitivity of the initial hypotheses to different aspects of the data. Visualization can help to understand how the initial hypotheses may be affected by data outliers, distribution shapes, or the presence of specific subgroups. btw do cascade domain knowledge with the insights from visualization which can guide for hypothesis refinement.

Visuals play a pivotal role in hypothesis evaluation - it could offer a clear &accessible means of presenting the results. Explore charts, graphs, and plots which illustrate the patterns/relationships identified during data exploration. Check for potential confounding variables that could impact the validity of the hypotheses. Visuals can help to identify covariates or factors that should be considered in the analysis Also, I would suggest, beyond statistical significance, do consider the practical significance of findings. Visualization can help to gauge the magnitude of effects. for example, Box Plots can display the "spread" & "central tendencies" of variables, aiding in assessment of the practical impact of variables on the hypotheses