Serial charts: categories, grouped values and split by field

One question, one chart. Picture the seven thirty briefing at a city water utility. The operations manager asks one question: which pipe material had the most main breaks this month? The answer is in the data, spread across four hundred break records in a hosted feature layer. Nobody wants to scroll a table while the director waits. A single bar chart answers it in two seconds: cast iron towers over ductile iron and PVC. So how do you build that chart in ArcGIS Dashboards, and how do you add a second layer of detail, like which breaks happened in winter? That's what this lesson is about. The tool is the serial chart, and once you understand its three settings, you'll use it on almost every dashboard you build.

What you will learn. By the end of this lesson you'll be able to do three things. First, choose the right way to feed data into a serial chart, using one of its three data options: grouped values, features or fields. Second, pick the statistic that matches your question, because counting records and adding up a number give very different answers. Third, use split by field to break each bar into series, so you can compare service request types across districts or outage causes across regions. We'll work through examples from service requests, power outages and water main breaks, and we'll check the logic with a few lines of Python along the way.

What a serial chart is. A serial chart is the dashboard element for any chart where values sit in a row, one after another, along an axis. The categories go along one axis, such as request type or month, and the numbers go along the other. You can draw them as bars, lines or filled areas. In ArcGIS Dashboards you add one from the element menu and point it at a layer. That's usually a hosted feature layer, or a Living Atlas layer, one of Esri's ready-made public layers. The big advantage over a static chart is that it's live. When someone filters the dashboard, zooms the map or selects features, the chart can recalculate on the spot, so the numbers always match what's on the map.

Three data options. Every serial chart starts with a choice of data option, and it decides what a bar means. Grouped values is the one you'll use most. You pick a category field, such as request type, and the chart makes one bar for each unique value, summarizing all the features that share it. Features is different: each feature becomes its own bar, which suits a short layer like ten pump stations, each with its own flow reading. Fields turns columns into bars, for example a storm response layer with separate fields for trees down, roads closed and shelters open. If you remember one rule, remember this: grouped values summarizes many records, while features plots them one by one.

Worked example: requests by type. Let's build one. The layer is this week's 311 service requests, one feature per call. In the data options we choose grouped values, set the category field to request type, and set the statistic to count. The dashboard counts the features in each group and draws five request types as bars. By default the bars come in alphabetical order, which hides the story, so we sort by value, descending, and now they're sorted largest first. With forty eight requests, potholes clearly lead, followed by streetlights out at thirty one. That's a chart a public works supervisor can act on before the coffee is finished.

Pick the statistic that fits. Count is only one of the statistics. Sum adds up a numeric field, so for a utility outage layer, summing the customers affected field tells you how many people lost power, not how many outage records exist. Average, or mean, is the total divided by the number of records, the typical value, like the average hours to restore power per region. Minimum and maximum show the extremes. Standard deviation is a measure of spread: think of it as how far values usually wander from the average. A small one means crews restore power in similar times; a big one means some outages drag on. Always ask what number the viewer needs before you pick one.

Split by field, checked in Python. Split by field adds a second grouping. The category field still makes the groups along the axis, but a split field breaks each group into separate series, one per unique value. Here are eight outage records for a small utility. We group by district, split by cause, and sum customers affected. In the dashboard that's grouped values, category district, split by cause, statistic sum. In pandas it's a pivot table, which is just a grid with categories down the side and series across the top. The first block lists the eight records. The pivot table line groups by district, splits by cause and sums customers, and print shows that grid on screen. You don't need Python to build the chart; it's only here to check the numbers. The result matches what the chart will draw: South's storm bar reaches eleven hundred ninety customers, while East's biggest problem was storms at five hundred forty, with equipment at two hundred.

Reading split series. Once you split, you choose how the series share space. Side by side puts storm, equipment and animal bars next to each other, which is best when you want to compare causes within one district. Stacked piles them into one bar, so the full height is the district total and the colored segments show what it's made of. One hundred percent stacked makes every bar the same height and shows shares, handy when districts differ a lot in size. Which would the storm response lead want? Probably stacked, because the first question is where the most customers are out, and the second is why.

Setting it up, step by step. Here's the order that works every time. Add a serial chart element and pick the layer, and if the layer has a date or status field, consider adding a filter here, such as open requests only. Set the data option, then the category field, then the statistic and its field. Add a split by field only if it answers a real follow up question. Next, sort by value, give each series a clear color, and write axis titles in plain words, like customers out, not cust_aff. Finally, test it: apply a dashboard filter or pan the map and watch the bars change. If they don't move, check the actions on the map or selector that should filter this chart.

Common mistake: too many series. The most common mistake is splitting by a field with too many unique values. Split traffic incidents by street name and you get forty colors per bar and a legend longer than the chart. Crew ID and asset ID cause the same mess. A quieter version of this trap is dirty data: if half the records say Storm and half say storm damage, the chart shows two series for one cause, and the storm total looks smaller than it is. The fix is to split only by a short, clean field, ideally one with a coded value domain, which is a fixed list of allowed values set on the layer. If the field is messy, clean it before you chart it.

Recap. Let's recap. First, the data option decides what a bar means: grouped values summarizes many records into one bar per category, features plots each record, and fields turns columns into bars. Second, the statistic has to match the question: count for how many records, sum for totals like customers affected, average for typical values. Third, split by field turns one bar into several series, which you can show side by side or stacked, but only split by a short, clean field or the chart becomes unreadable. Back at the water utility, split the pipe material chart by season, and the winter breaks stand out on every bar. With these three settings, one serial chart can answer the question in the briefing room before anyone asks for the spreadsheet.

Your exercise. Now it's your turn. Open ArcGIS Online and find a layer of service requests or incidents, either your own or a public one from Living Atlas or a city open data site. Create a dashboard and add a serial chart using grouped values, with request type as the category and count as the statistic, sorted largest first. Then add a split by a short field such as status or priority. Try it side by side, then stacked, and keep whichever one answers the question faster. Write one sentence your manager could read from it. Next lesson, we'll put time on the axis and add trend lines.