Correlations are one of the most powerful features of the app, allowing you to uncover meaningful connections between different parameters in your trackers. Properly understanding and configuring correlations can help you gain deeper insights into your habits, progress, and overall data trends
What is a Correlation?
A correlation measures the relationship between two parameters – a Source and a Target – based on the pairs of their recorded values. The strength and direction of this relationship can provide valuable insights into how one parameter might influence another.
Note 2: The correlation methods used in this app primarily detect linear relationships, meaning they capture direct, proportional changes between parameters (the more A – the more B, the less A – the more B etc.). If the connection between your parameters is non-linear, the calculated correlation might not fully capture the true nature of the relationship. You can always take a look at the charts to observe it more closely
Note 3: The correlation value (and thus the correlation strength) doesn’t necessarily reflect how much one parameter affects the other, but rather how tightly their values move together. To better understand the nature of the connection, check out the charts for a more detailed view of the data distribution
Note 4: Correlation coefficients do not account for the order of measurements or the dynamic of changes over time. They simply measure the overall association between two sets of values, without considering the direction or speed of changes. For insights into the timing and sequence of your data, consider exploring trends visually using charts
How Correlations are Calculated
Data Pairs
- Correlations rely on pairs of values, with the Source parameter providing the first value and the Target parameter providing the second
- For parameters in the same tracker, data pairs are simply taken from the same samples, as long as neither parameter is skipped
- For cross-tracker correlations, daily aggregation is required, as parameters do not share samples. In this case, values are grouped on a daily basis before pairing
- Example 1 (Same-Tracker): If you have a fitness tracker with parameters like Push-ups and Calories Burned, each sample might record the count of push-ups and calories burned in a single workout session. These data pairs naturally align within each sample
- Example 2 (Cross-Tracker): If you track Workout Intensity in one tracker and Mood Level in another, the app can only pair these values on a daily basis, requiring aggregation of multiple samples if you have more than one sample per day
Daily Aggregation
- When correlating parameters from different trackers, or if you want to analyze trends over time, you need to aggregate your data on a daily basis
- You can choose from several aggregation methods for both the source and target parameter, including:
- Sum – Total of all values for the day
- Average – Mean of all values for the day
- Last – The last value recorded on that day
- First – The first value recorded on that day
- Min – The smallest value recorded on that day
- Max – The largest value recorded on that day
- Daily aggregation can also be applied to same-tracker correlations if you want to smooth out or summarize daily trends, but it is required for cross-tracker correlations
- Example: If you track your Steps and Calories Burned separately, and record multiple samples throughout the day, the app will need to aggregate these values before calculating the daily correlation
Daily Offset
- You can apply a daily offset to explore time-shifted relationships. This feature is particularly useful if you suspect that one parameter might affect another over time
- For example, if you want to know how today’s workout impacts your Mood Level tomorrow (offset = 1) or in four days (offset = 4), you can set the appropriate offset to capture this potential delay
- The app supports up to 7 days of offset, allowing you to explore a wide range of potential causal relationships
- Advice: When using offsets, be mindful of how you set the source and target parameters. The source should be the parameter you believe can influence the target over time, not the other way around
Correlation Types
- The app supports several correlation functions, each suited to different types of data:
- Phi – For binary data (e.g., Rained (Yes/No), Indoors (True/False))
- Pearson – Primarily for numerical data with a linear relationship (e.g., Height vs. Weight, Hours Studied vs. Test Scores)
- Spearman – Primarily for ordinal data or when you expect a monotonic relationship (e.g., Mood (1-5) vs. Energy Level (1-10))
- Point-Biserial Correlation – For binary and numerical data (e.g., Success (Yes/No) vs. Time Spent (minutes))
- Rank-Biserial Correlation – For ordinal and binary data (e.g., Pain Level (1-10) vs. Medication Taken (Yes/No))
- Note: Some correlation types can be applied to different data types, so we try to calculate all that we can to provide you with more insights. Nevertheless, understanding the nature of each correlation type (e.g., assumptions concerning data distribution) may be useful in your investigations
- You can also choose Best Mode, which automatically selects the correlation type with the highest absolute value for your data
- Advice: Use Best Mode if you’re not sure which function to choose, but if you have a specific expectation about the nature of the relationship, pick the most appropriate function directly
Interpreting Correlation Results
- Correlation Value (-1 to 1) – Indicates the direction and strength of the relationship:
- Positive: As the source increases, the target tends to increase
- Negative: As the source increases, the target tends to decrease
- Correlation Strength – Shows the closeness of the relationship:
- Negligible (0 – 0.3)
- Weak (0.3 – 0.5)
- Medium (0.5 – 0.7)
- Strong (0.7 – 1)
- Correlation Significance – Indicates the reliability of the correlation:
- Calculated using Student’s T-test
- Levels: Insignificant (p > 0.05), Low (p < 0.05), Medium (p < 0.01), High (p < 0.001)
- The p-value quantifies the likelihood that your observed correlation happened by random chance
- Larger data sets and stronger correlations tend to have higher significance
- Advice: Pay close attention to High or sometimes Medium significance levels, as these are more likely to reflect meaningful relationships
- Correlation Direction – Simply the sign (+/-) of the correlation value
- Pair Count – The number of data pairs used to calculate the correlation, which can impact the reliability of the result
Saving and Revisiting Correlations
- You can save interesting correlation configurations to Favorites for quick access in the future
- Keep in mind that as you add or modify samples, the correlation results may change over time, reflecting your evolving data
- Advice: If you find an interesting correlation, consider saving it to track how it evolves as your data set grows
Correlations can be incredibly powerful for revealing hidden insights in your data. Take the time to experiment with different configurations, offsets, and parameter pairs to unlock deeper understanding and make the most of your tracking journey!