How can you ensure your R&D team interprets data correctly?

It is essential to make certain that the team comprehends the potential implications of their findings and adheres unwaveringly to ethical standards in their work. Establishing a supportive environment that fosters precise data interpretation within the R&D team is pivotal. Consistent communication, continuous training, and a steadfast commitment to quality will collectively enhance the team's ability to interpret data accurately, thus significantly contributing to the overall success of research endeavors.

While defining goals is important, I believe asking the “right” questions and being open to any answer that presents itself through data are even more critical to successful interpretations. By “right” questions, I’m referring to unbiased, open ended, and objective. The “wrong” questions will inherently lead the team to inaccurate, biased, and potentially altogether incorrect interpretations. In many cases, it takes an outsider view or new eyes to truly help distinguish between right and wrong, biased and unbiased, objective and subjective. Seek fresh eyes!

1. Always keep in mind what the end customer needs and if the data you are using ultimately affects the way the product is used by them. It isneasy to get bogged down with details that do not matter in the overall purpose. 2. Ensure you have sufficient expert interpretation to incorporate user variability into your analysis. 3. Finally have a clear and representative training and validation set.

It is quite simple a) experimental design, b) using controls c) use relevant models. -experimental design is critical component that is to say the parameters one measures need to be relevant to your next decision inflex point Control needs to be robust relevant and above reproducible. Relevant models it is plural for a reason a single model rarely provides enough data to confirm further pursuit

Data accuracy is an important aspect of data quality that affects how reliable and useful the data is for various purposes. The R & D team needs to ensure the following factors for correct data interpretation: 1. Define clear and consistent data standards and rules for data entry, storage, and analysis. 2. Conduct data audits or audit trails to verify the accuracy and completeness of the data. 3. Implement data quality checks and validation methods to identify and correct errors, inconsistencies, and outliers in the data. 4. Document and report any changes or issues related to the data accuracy and quality.

Quality of the data is critical. One should know what are the sources of variation of the measurements and how they are controlled (if they are). Then take them into account in the analysis (blocking, random effects or covariates). This should be considered beforehand to the production of data, but during data collection all relevant observation impacting data quality must be collected as well. This will allow to explain variability and explain outliers.

Understanding the quality of the data and data collection methods is critical. Researchers often aren’t aware of all the sources of noise that are impacting their experiment. Confounding factors may indicate a response that would actually be attributed to a different factor or interaction. Additionally, measurement technique can have major impact on results and the uncertainty. A good rule of thumb is don’t start running costly experiments until you intimately understand your measurement system.

Even before understanding where your data comes from, you need to to establish the 'Class' of data required. Very simplistically you can categorise them into 3 classes:- Class 1 - Observable not easily collectible - e.g information on fairly upset customers at a hotel concierge who do not send written complaints. Some 'ground-level' human sensing is required. Class 2 - non-exhaustively collected data - e.g survey data that typically don't go in as input into the Machine Learning models you develop Class 3 - those that are available in digital format Once you develop a feel for the 'Class' of data that you require then you can think about how 'data representable' your problem is and subsequently where you data sources come from!

It is so important to remember the basics! Two decades ago, we embarked on HTS journey under the premises that with enough data would self-assemble and give us the answer to a challenging problem. But what we really needed to do was develop a hypothesis, test it ; observe the data and then draw a conclusion. Sometimes the simplest principle eludes us because they seem inconvenient.

When one has to deep dive into some search, that search has to be initiated with a public domain website. as public domain sites such as PubMed, Cochrain Review, etc. have access to all the full articles and have the maximum number of published articles. Keywords for the search play an essential role, as the right search will fetch you the right information. The articles that have limited access or data can be used for cross-referencing. The number of references in the same articles can give the right direction for a search.

Data is like a cloud in front of us. In order to be transformed into understandable and meaningful information, it needs to be acquired by the right measurement techniques and devices. Therefore, initially, how data will be collected is crucial. Moreover, if the data has been collected previously, how it will be interpreted becomes another important issue. We actually name this process as analysis. Meaningful expressions and interpretations can be derived from data that has been collected or analyzed in this way. Finally, working with skillful people who are experts in their fields is another of the most important criteria to achieve correct results by implementing the right methods and approaches.

Incorporation of Internal/external negative and/or positive control ensures the robustness of the techniques used for data analysis.

Test data analysis is a very deep subject, and in my expirience, for a good test execution, people dooing the test should be given a very clear and standardized checking tool - that provides a "fail"/"pass" answer - with a "gray area in the midle. Tests taht fall into thos category shoud then be avaluayed by an expert for that particular testing subject.

Choosing appropriate method is very critical as - It may directly impact on your results. -Wrong results can mislead you and the final goal cannot be achieved. - Using proper method and techniques can save the valuable money as well time.

Match methods to questions: Don't apply fancy algorithms to simple questions. Choose methods that address your specific research goals. Seek expert guidance: Consult statisticians or data scientists for complex analyses, ensuring proper tool selection and interpretation. Validate your choices: Test and compare different analysis methods to ensure the chosen approach yields reliable results.

Visualisation and communication of findings is important because it can save you from big troubles that can come if the findings are not shared within time.

Visualisation is the first step of data analysis. Even before cleaning. When on of our researchers come to me with data we start by having a look at the data with graphic tools. When it comes to presenting results it is important to keep in mind that the statistician must keep factual and not biaised the outcomes with representations that influence the interpretation (zooming with axis settings, no error bars).

Visualisation = communication Therefore good data visualisation starts with understanding the public. * What knowledge do they have of various types of visualisations? And what knowledge do they have of the data? * Are there e.g. societal factors about colors with specific associations? * Use an visualisation that is the simplest representation of your message and test this with your public. All extra information that attracts attention, weakens your main message. And last, but not least: Make your visualisations digital accessible for all users, by using correct colors, contrasts with clear descriptions.

Clarity over complexity: Prioritize clear and concise visualizations that effectively communicate key insights to diverse audiences. Storytelling with data: Don't just show numbers; weave a narrative, highlighting trends, patterns, and actionable takeaways. Context matters: Provide context for visualizations, including data limitations, error bars, and clear labeling for informed interpretation.

We have two eyes and ears for a reason. Visualization the most important part of sharing the data such that your team will understand and use the data effectively. The presentation and verbal explanation of the data is critical to the digestion of the information by your colleagues. When presenting data, I rarely have anything more than a title accompanying my visual data results. I want my colleagues to see the visual results while listening to me explain them. The visual connection to the verbal explanation is far more successful than someone reading the explanation beside the visual.

Validate data interpretations by cross-referencing multiple sources, employing diverse analytical approaches, and conducting peer reviews within the R&D team. Challenge assumptions and conclusions by actively seeking contradictory evidence, encouraging debates, and exploring alternative explanations. Foster an environment that encourages critical thinking, inviting perspectives from multidisciplinary experts. Employ robust validation methods, such as sensitivity analysis or simulation, to test the reliability and robustness of interpretations. Emphasize the importance of rigorous scrutiny and openness to refining conclusions, ensuring a comprehensive and reliable understanding of the data for informed decision-making within the R&D team.

This is where many ignore the effectiveness of multiple experiments/tests (and reproducibility of same) for multiple data points before one can rely on the results to be interpreted. Interpretation of a 1-time fluke test result does more harm than good.

In the process of interpreting your data you should be able to validate and verify the out come. The method and tools you are using no matter how well known or popular they are you have to verify them according to your parameters and practice and run sone previously worked on data

I think, that before the Digital Transformation Era, validation of the data an its source was extremely crucial. Even nowadays, data cleaning is crucial to avoid GIGO (garbage in - garbage out) situation before doing the interpretation of the data collected. Thanks to ML and GPT-models scraping and formatting tasks become easier and much-much faster.

La forma de interpretar los datos debe considerar la implicancia que tiene cada resultado del análisis, y si es posible el alcance en aspectos técnicos económicos y ambientales. Por ejemplo la interpretación estadísticas de aspectos técnicos de una opción puede sugerir que una opción sea aceptada pero si esta opción es económicamente inviable entonces de seguro la decisión de adoptar es cuestionable

Interlinking of dots is very important, reflect on past projects, assess the effectiveness of methodologies used, and identifying areas for improvement. Discussions on lessons learned from the past will always improve the progress of the current project.

Leverage data experiences to drive continuous improvement within the R&D team. Encourage reflection on past methodologies, successes, and failures, extracting lessons for enhanced future practices. Foster a culture of learning by documenting insights, establishing knowledge-sharing sessions, and integrating feedback loops for iterative advancements. Emphasize skill development in data analysis, interpretation, and application. Implement agile methodologies to adapt swiftly to emerging trends or challenges. Embrace innovation, experimenting with new techniques and technologies to refine data-driven strategies.

All engineering or scientific work shall be defined with a clear and concise glossary of terms, and include any units and proper designation for a proper and uniform results evaluation

Continuous improvement in data interpretation is vital. Reflecting on past data experiences allows for recognizing successful approaches and areas needing improvement. Learning from mistakes and failures is crucial in research and development. They offer invaluable lessons to refine data skills and knowledge. Embracing these experiences as learning opportunities strengthens the team's capabilities in handling diverse datasets. Staying updated with the latest trends in data science and technology is key. For an R&D section, this perspective is gold.

Document data analysis processes: Maintain detailed records of methods, tools, and rationale for future reference and replication. Debrief and iterate: Regularly discuss what worked well and what could be improved in future data analysis practices. Stay updated: Encourage continuous learning of new data analysis techniques and tools to adapt to evolving needs and challenges.

One thing I have found helpful to ensure my team interpret data correct is to give them repat mission which they have done previously as new one or give similar mission to my people seperately (tell them it's a secret test) and see the results. Another way is to do some test by myself in absence of team and comparing the results. These steps doesn't mean lack of confidence on team but Quality assurance is required every stage of work.

Having people taht become experts on test evaluation is mandatory, but finding them is a big task, because they can not be brought in from another company and be effective, due to different softwares, testing methods and strategies. It is far better to "groom" someone with existing testing experience, that has shown interest in a particular subject. That person can become verry effective realtivly quickly and give mutch better results .

I find it useful when I assess data generated from previous experiments while deciphering possible confounding factors that could impact on the data. Data interpretation can be complex especially in generating hypothesis but can become a valuable tool in avoiding worthless chase of non useful projects. We must strive to save time and costs but it all boils down to clear interpretation of generated data!

Self-delusion is deadly. “You must not fool yourself, and you are the easiest person to fool.” (Feynman). Bounce off ideas with teammates, friends, aliens. Review, adjust, correct, and review some more with the above "personas". E.T. is perfect, as that creature always seem to be asking questions. Avoid people that pretend having read and carefully reviewed your stuff: "Yeah, yeah, it's all good". Enjoy the ride!

On peut apprendre plein de méthodes et outils, le meilleur moyen d'y arriver est d'apprendre aux équipes à prendre du recul, être curieux pour comprendre d'autres manières de penser. C'est peut être le plus dur mais probablement le plus efficace à terme.