Reading a trend
When your rows are in order (months, weeks, years), the explanation panel adds a few lines about change. Here is what each one is, worked on the sales template.
First to last
The last value compared with the first: 4,210 in January to 6,730 in October is a rise of 2,520, or 60%. It ignores everything in between, so a good first month and a bad last month give a "fall" even if the middle was strong. Read it together with the highest and lowest.
Growth per step
The steady percentage that would take the first value to the last over the same number of steps (the compound growth rate). 4,210 to 6,730 in nine steps is 5.4% per month. It's a summary, not what happened each month; the biggest single step was April to May.
The trend line and "per step"
The dashed line is the straight line that sits closest to all the points (least squares). Its slope is the "about 309 per step" in the summary. Turn it on with the trend-line box in the maker. It shows direction; it doesn't know about seasons, price changes or the month you were closed.
R², or how well the line fits
A number from 0 to 1. Near 1, the points sit close to the line and "rising" means something. Under 0.35 the points scatter, the line is a weak summary and the panel says so. Sales at 0.62 is in between: the direction is real, the month-to-month path is bumpy.
Next step, if the trend holds
The line extended one more step. It's an arithmetic extension, not a forecast. It assumes next month behaves like the straight-line average of the past months, which December rarely does. Use it to ask "is that plausible?", not to plan stock.
When there's no trend to read
- Fewer than four points. Two numbers always make a line; three can go anywhere next. The panel says so.
- Categories instead of a sequence. Scores by student have no "first to last". The panel detects month, day, week, quarter and year labels; anything else it treats as categories and skips the trend lines.
- A single big jump. One outlier can tilt the whole line. The panel names rows that stand out from the rest: values more than 1.5 times the interquartile range beyond the middle half of the data, the usual box-plot rule. It needs at least five rows.