Key points for time-series comparisons: 7 perspectives useful in data analysis

In data analysis settings, time-series comparisons are indispensable for understanding changes over time. By comparing and analyzing time-series data such as sales, page views, or sensor measurements, you can gain many insights, including trend detection, anomaly detection, and evaluation of the effects of initiatives. That said, if you compare data carelessly while ignoring seasonal fluctuations or external factors, you may reach incorrect conclusions. For example, directly comparing sales in busy and slow seasons might lead you to mistake a seasonal downturn that should have been predictable for an abnormality. Therefore, to compare time-series data correctly and extract useful insights, it is important to keep several key perspectives in mind.

This article explains seven perspectives that are helpful when comparing and analyzing time-series data. By understanding and applying these points, you can correctly interpret data changes and use them to support decision-making in business and research.

Table of contents

• Perspective 1: Comparison with the past (e.g., year-over-year)

• Perspective 2: Comparison with targets and forecasts

• Perspective 3: Consideration of seasonality and periodicity

• Perspective 4: Outlier detection and impact analysis

• Perspective 5: Comparison with other entities and overall trends

• Perspective 6: Factor decomposition and use of multiple data sources

• Perspective 7: Appropriate visualization and choice of analysis granularity

• Conclusion

Perspective 1: Comparison with the past (e.g., year-over-year)

A basic approach in time-series data analysis is comparing with past values. By comparing current data with past reference points such as the same month last year or the previous observation, you can grasp trends in increases or decreases and growth rates. In this perspective, it is important not only to look at simple numerical differences but also to check change rates (for example, a year-over-year change of 〇%). Looking at change rates makes it easier to capture relative growth or decline regardless of absolute values.

When comparing with the past, setting an appropriate reference point is key. Even for the same metric, the trend you observe can differ greatly depending on the period you choose for comparison. For example, when comparing sales to the previous year, comparing with the same month or quarter one year ago helps align seasonal factors. For weekly data, comparing with the same week in the previous year rather than the previous week can remove the effects of the arrangement of weekdays and capture more accurate trends. Choosing the wrong reference point can lead to misreading the true trend, so be careful when selecting comparison points.

Long-term comparisons with the past are also useful. By lining up data from several past years and seeing where the current value sits within the long-term trend, you can grasp the bigger picture without being misled by temporary fluctuations. For example, comparing ten years of industry-wide sales data can help you interpret business cycles or market growth pace. In Perspective 1, it is important to properly choose past reference points and evaluate change rates to accurately assess upward or downward trends. Note that you may also compare consecutive periods such as month-over-month or quarter-over-quarter to understand recent movements, but even then you should interpret the results with seasonal differences in mind.

Perspective 2: Comparison with targets and forecasts

Another important perspective is comparison with target values and forecasts. By analyzing how actual data compares to preset targets or forecasts, you can evaluate plan achievement and the accuracy of forecasting models. In this perspective, it is key not only to see whether a target was met or missed, but also to investigate the size of the deviation and its causes.

For example, if a sales target of 100 was set at the beginning of the month and the month-end result is 90, the target was not reached. At that time, it is important to choose an expression such as a 90% achievement rate relative to the target or a gap of -10. The achievement rate is useful for intuitively grasping overall progress, while the absolute gap helps understand the scale of shortfall or excess. Which metric to use should be chosen according to the purpose of the analysis and the audience (bosses or stakeholders).

Comparisons with forecasts can also be used to validate forecasting model accuracy. For example, suppose you used a machine learning model to forecast weekly website traffic. Comparing actual values with forecasted values and analyzing discrepancies can reveal seasonal factors or unexpected events the model failed to capture. If there are weeks when forecasts are significantly off, investigating their causes can lead to improvements in future predictions.

Furthermore, comparing targets and results is not only for outcome evaluation but also an opportunity to verify the appropriateness of target setting. If targets are consistently missed, the plan values may be too ambitious or the strategy flawed; conversely, if targets are easily exceeded, they may be too conservative. Regularly reviewing and feeding back findings into the next planning cycle is important.

Thus, Perspective 2 provides insights for both plan management and model validation by comparing actuals with targets and forecasts.

Perspective 3: Consideration of seasonality and periodicity

Seasonality and periodicity are indispensable perspectives when analyzing time-series data. Many datasets exhibit seasonal fluctuations over the year, weekly patterns, or even multi-year business cycles. If you compare data without considering these patterns, you may be misled by apparent differences.

A typical example is products whose demand differs greatly between summer and winter. It is natural for air conditioner sales to increase in summer and decrease in winter. In this case, directly comparing July sales with October and concluding that "sales have decreased" is inappropriate. Instead, compare with the same month in the previous year or use seasonally adjusted values to capture the pure trend change. Similarly, when handling weekly sales data in industries where weekday and weekend patterns differ, make efforts to align weekday effects by comparing weekends with weekends, for example.

Also pay attention to long-term cycles that span years. If data are affected by business cycles or trends that repeat every few years, short-term comparisons alone may miss these signs. For instance, if an economic indicator rises and falls in a 3–5 year cycle, it is important to judge the current value in light of that cycle. Analyzing past periodicity can offer hints for anticipating upcoming changes.

In Perspective 3, keep time-specific patterns such as seasonality and periodicity in mind and perform adjustments or corrections as needed when comparing data. Using seasonally adjusted data or aligning comparison periods like year-over-year comparisons makes it easier to identify true trends and anomalies.

Perspective 4: Outlier detection and impact analysis

Time-series data may contain outliers (anomalous values) that deviate greatly from the overall trend. Outliers often arise from special events or data errors; they can themselves be important discoveries, but they require caution during time-series comparisons.

First, detect outliers before conducting comparisons. Plot graphs to visually check for obvious deviations, or use statistical methods to identify anomalies. For example, if an indicator normally varies by about ±5% per day but spikes by 50% on a certain day, that should be flagged as an outlier. You need to determine whether that spike was caused by a temporary campaign, a data aggregation error, or something else.

Once you discover outliers, perform impact analysis. In other words, compare trends when including the outlier versus excluding it. For instance, when calculating monthly averages or totals, a single extreme value might disproportionately raise the overall result; be cautious when evaluating that month. Calculating values both with and without outliers helps you understand typical performance.

Investigating the cause of the outlier itself is also important. If the anomaly was due to a temporary factor, it may not recur, but if it signals the start of a new trend (a paradigm shift), it should not be overlooked. Rather than dismissing it as mere noise, investigate the background to gain insights useful for future strategy. Perspective 4 requires correctly handling outliers and controlling their impact to perform more accurate comparisons and analyses.

Perspective 5: Comparison with other entities and overall trends

In addition to analyzing a single time series, comparing it with other entities or the overall trend deepens your understanding of the data’s significance. Comparing your data with competitors’ movements, market averages, or data from other regions or departments can reveal whether your situation is relatively good or bad and what characteristics stand out.

For example, suppose your sales increased by 5% year-over-year. At first glance this looks like growth, but if the market as a whole grew by 10% during the same period, your growth rate is below the market average and you may be losing market share. Conversely, if the market is flat or slightly declining while your company grows by 5%, that would be an impressive performance. Comparing with overall levels is essential to understand relative positioning that cannot be grasped by looking at your own data alone.

Comparing multiple categories or regions of the same time series can also yield useful insights. For instance, comparing monthly sales trends of chain store A and store B might reveal different growth patterns between a store reliant on repeat customers and a store that attracts many new customers. In such cases, look not only at total sales but also at sales composition, number of customers, and average transaction value to clarify reasons for differences. When handling multiple entities, ranking comparisons can be useful as well. For example, ordering all stores or products by sales highlights outstanding performers and underperformers. Identifying commonalities and differences among top and bottom performers helps extract success factors and issues.

Tracking changes in composition ratios relative to the whole is also important. Observing how each department’s share of total company sales changes over time reveals which departments are growth engines and which are shrinking relatively. Time-series comparisons of composition ratios can inform resource allocation and prioritization decisions.

Perspective 5 provides a relative evaluation by comparing your data with others. Rather than celebrating or worrying over changes in isolated figures, adopt a broader view to correctly interpret what the data imply.

Perspective 6: Factor decomposition and use of multiple data sources

When comparing time-series data, analyzing decomposed factors that make up the data or combining related datasets is effective. To reveal causes or backgrounds that are not apparent when tracking a single metric, break down the data or cross-reference it with other indicators.

A representative example is analyzing a sales time series. Sales can be decomposed into "number of customers × average spend per customer." Even if monthly sales appear flat compared to the previous month, the breakdown may show that customer count fell while average spend rose, resulting in an overall flat figure. In this case, you can interpret whether there is an issue with attracting customers or whether strategies to increase spending among existing customers have succeeded. Changes that are hidden when looking only at a single aggregate value can become clear through factor decomposition.

Combining multiple time-series datasets is also powerful. By comparing external factors with your internal data, you can investigate causal relationships and associations. For example, comparing temperature changes with beverage sales during the same period may reveal a correlation between rising temperatures and increased sales. Similarly, overlaying website traffic data with advertising spend time series can help estimate the impact of ads on traffic increases. Combining multiple datasets can highlight background factors that are difficult to judge from a single time series.

Perspective 6 emphasizes delving into the target data itself (breakdowns and decomposition) and cross-referencing surrounding data to achieve deeper comparative insights. In practical data analysis, such multifaceted approaches yield substantive insights beyond mere observations of increases or decreases.

Perspective 7: Appropriate visualization and choice of analysis granularity

Finally, consider data visualization and the choice of granularity (aggregation interval). No matter how good your analysis methods are, if the results are not presented intuitively, stakeholders will not grasp them. In time-series comparisons, selecting appropriate chart types and aggregation granularity is key to extracting insights.

The basics of visualization are to use charts that track time, such as line charts or bar charts. When comparing multiple time series, overlay multiple lines on a single chart or use specialized charts that show only differences to make contrasts clear. When comparing monthly trends across several years, arranging the same months side by side or overlaying yearly lines with different colors can be effective. Visual pattern recognition helps you notice changes that a table of numbers might miss.

However, be aware that the vertical axis (scale) can greatly change impressions. If you narrow the vertical axis too much to emphasize changes, small differences may appear as steep fluctuations. Conversely, too wide a range can bury important changes. Set axis scales and units appropriately according to the purpose of comparison to avoid misleading visualizations.

Setting analysis granularity is also important. What period you aggregate the data into greatly affects what you see. Comparing daily data as-is may distract you with small ups and downs and cause you to miss the main trend. On the other hand, annual data may be too coarse, hiding short-term variations and seasonal effects. Appropriate granularity depends on the analysis objective: for understanding overall flow, aggregate monthly or quarterly; for short-term effect verification, use daily or weekly granularity. Flexibly switching granularity according to needs is important.

When comparing series with very different scales, indexing can be useful. Converting values to an index where a reference period equals 100 allows direct comparison of relative changes even when absolute values differ. For example, plotting indexed sales amounts and customer counts lets you see at a glance which is experiencing larger fluctuations despite differing units. In Perspective 7, carefully designing effective presentation and aggregation methods helps make insights from time-series comparisons clearer and more accurate.

Conclusion

We introduced seven important perspectives for time-series comparison analysis. From comparisons with the past and with targets, to considering seasonality, handling outliers, relative comparisons with other data, factor decomposition, and visualization and granularity choices—covering these points comprehensively will significantly improve the quality of insights derived from data. These perspectives also serve as a checklist during analysis, helping to prevent hasty conclusions and missed important changes. For data analysts, improving the accuracy of time-series comparisons is a crucial skill that leads to accurate trend understanding, better forecasts, precise problem identification, and ultimately stronger business outcomes.

Also, do not overlook the quality of the data that underlies the analysis. Especially when handling advanced time-series data such as location information, positioning accuracy affects the reliability of results. For example, in position data analysis using GNSS (Global Navigation Satellite System), leveraging a high-precision positioning device LRTK (iPhone-mounted GNSS high-precision positioning device) that can be attached to a smartphone can improve the accuracy and consistency of the data. In scenarios requiring high-precision location data—such as surveying, monitoring, or obtaining location information from various IoT sensors—LRTK can be a powerful aid. With high-precision time-series data, analyses based on the perspectives discussed in this article will become even more reliable.

Put these time-series comparison points into practice and extract valuable insights from your data.