Wie aktualisieren und pflegen Sie Ihre Datenqualität im Laufe der Zeit?

Data quality should be a constant business as usual activity. Many businesses have treated data quality as a project or distinct piece of work to clean up data, without putting in place BAU processes to keep it clean at all times.

The first step in assessing data quality is to thoroughly study your data and identify the necessary data points. Eliminating outliers and inconsistent data should be the initial focus. I personally set predefined parameters to determine what is required and what is not, which greatly aids in the subsequent data cleaning process.

Assessing the data quality is the cornerstone of maintaining and updating data quality over time. Regularly evaluate the datasets to identify anomalies, missing values, or inconsistencies that may arise during data collection or integration processes. Utilize data profiling techniques and quality metrics to gauge the accuracy, completeness, and consistency of data. Establish clear criteria for data quality, aligning with the specific needs and objectives. A systematic assessment not only allows for the timely detection of issues but also informs targeted improvements and corrective actions, ensuring that data remains reliable and valuable in the long run.

I recommend to start with some simple analyses just to get a feeling for the data. Those graphs and reports are "throwaway reports". So you can plot data in different ways to see if there are outliers. You can create simple descriptive statistics like summing and counting data and see if it matches the original sample size. Check for duplicates and maybe even misspelled answers (in surveys etc.) which are actually all the same data. This would require data cleanups. Regex could be a tool-candidate to check for variances in text outputs.

Maintaining and updating data quality over time involves a systematic and continuous approach. Establishing clear data quality standards and conducting regular audits helps in identifying and correcting errors. Utilizing automated tools for data validation and cleansing can enhance efficiency and accuracy. It's also important to cultivate a culture of data responsibility among team members. Adapting data governance policies to accommodate evolving data types and sources is crucial. Regular training for the data team ensures they stay updated with the latest technologies and practices. This comprehensive strategy ensures sustained high-quality data, facilitating reliable analysis and decision-making.

Always consider using 3rd party data sources when cleaning and validating personal data. Whilst some methods can clean according to "rules", using 3rd party data is the most effective for personal data accuracy.

Talking in the scope of Excel, you can use Power Query or simple functions within a worksheet. For data cleaning, Excel offers a variety of tools like checking for wrong data types or using the "remove duplicate" function. I strongly recommend having a good understanding of your dataset, before you blindly remove duplicates. In doubt, ask the person who provided the data, pulled it via e.g. SQL and the like and adapt if errors occur. As a data analyst you need to have a good understand of the whole ETL process (Extract, Transform, Load).

You've got your raw data, and it's kinda messy, maybe a bit all over the place. What you gotta do is dive in with a mix of techniques: cleansing to get rid of the junk, transforming to shape it up right, standardizing to keep things consistent, deduplicating because nobody likes repeats, enriching to add that extra zing, and imputing to fill in the blanks. It's like giving your data a total makeover, turning it from rough and unruly into something sleek, usable, and, most importantly, reliable for whatever analysis or model you're running. This step, though sometimes tedious, totally makes or breaks your results. Be careful! This stage is really important.

Assess the consistency of your data across different sources, systems, or time periods. Look for any discrepancies or contradictions in the data. For example, if you have different records for the same entity, ensure that the data is aligned and standardized.

Proceed to the crucial phase of data cleaning after identifying quality issues. Employ diverse methods and tools, including data cleansing, transformation, standardization, deduplication, and imputation. Correct, modify, delete, or add data to enhance its quality and usability. Data cleaning is the pivotal bridge between data assessment and meaningful analysis, ensuring that your dataset is refined, accurate, and primed for extracting valuable insights with confidence.

Documentation is key. Always work in a way that 3rd parties could work with your analyses and data independently. Tools too use: - plain text editors as "readme" files - company wikis - in Excel: separate worksheets, notes within cells which act like a mouseover description

Evaluate whether your data is up-to-date and relevant for your analysis or decision-making needs. Assess if there are any delays or gaps in data collection or updates. Outdated data may not accurately reflect the current situation, potentially leading to misleading insights or decisions.

Documenting data is indeed a critical step in managing and leveraging data effectively within an organization. By creating and maintaining metadata, data dictionaries, data catalogues, or data lineage records, you establish a structured framework that describes various aspects of your data, such as sources, structures, formats, definitions, relationships, and processes.

Clear documentation helps users comprehend the data's context, structure, and usage. Proper documentation ensures data accuracy and consistency over time. Maintaining SOP on Data sources, process of hygiene, and developing comprehensive data dictionaries which describe the elements and attributes. As a Process documenting the process, will make sure even a person changes process will have its impact .

If you use a data transformation tool - such as dbt or dataform -, you can add documentation to your models and make it available with a simple data lineage diagram

After defining exactly what is important to monitor, I would recommend using a data quality tool (greatexpectations, soda and similars) to automate data quality checks. Data transformation tools (again, dbt and dataform) also have integrated assertions to make sure data is being tested from the beginning. With tests being ran, you'll need to make sure they generate accessible data that can be used in dashboards and alerts.

I worked on a customer project, who collected data through forms in Google sheets. We set up a main dashboard, which aggregated the results. It does not have to be fancy, and you can monitor with simple tools and functions main insights about your collected data. There is no right and wrong about the approach, it just needs to be beneficial to the corresponding project.

Guardians of Data Vigilance Data quality isn’t a one-time quest; it’s an ongoing watch. Set up alerts, track anomalies, and patrol for inconsistencies. Like vigilant sentinels, we safeguard against drifts and glitches. Remember, data quality is our fortress—protect it with unwavering diligence!

If it is online data, which is continuously flowing, like data from a sensor in the case of predictive maintenance, then monitoring it is crucial. There are chances for the data to change (data drift) or to produce erroneous values. In such cases, monitoring the data is better to find out potential issues.

- Establish regular data audits and reviews to continuously assess and ensure the adherence to data quality standards. Setting up automated alerts for anomalies or deviations from data norms can help identify issues early. 🚨 - It’s crucial to have a clear data governance framework that defines roles and responsibilities for data quality management. 🔍 - Integrate data quality tools that offer features like data profiling, monitoring, and cleansing into your systems. 🛠️ - Regular training and updates for your team on best practices in data handling and compliance with new data protection regulations (like GDPR or CCPA) are essential to keeping your data quality high as standards and technologies evolve. 📚

Once data audits are on track, and documentation is in progress, need to implement rules to enforce the data integrity. To improve the data quality routine checks are must and for this need to enable automated monitoring systems and most important should work on feedback mechanism by users input.

Given a proper way to monitor your data inflow, you can adjust for possible errors on the fly. Once you recognize discrepancies in survey data streams, you should act fast and fix for examples survey forms. This should also be well documented and respected in the following data analyses. The one, who is analyzing data needs to be aware of time stamps, when changes in the data stream occurred. You may split your data set and select proper samples to run valid analyses.

Implement data quality control- such as data quality policy, standard, guidelines these are the rules and procedure that you apply to your data source, processes and outputs to ensure and improve data quality.

Conclude your data quality journey by instituting periodic and systematic reviews. Utilize tools and techniques such as data quality audits, feedback sessions, surveys, or benchmarks to continually assess and verify data quality status and progress. This ongoing process ensures that your data quality efforts remain effective and adaptive. Regular reviews enable validation, confirmation, and improvement of data quality results and achievements, maintaining the reliability and integrity of your dataset over time.

One can consider using a feature store for online data. Feature store is like a warehouse of features for online machine learning pipelines. This tool has the feature of detecting concept and data drift issues, which can be handy while working on online machine learning.