Study suggests that judgmental forecasting of trends in time-series data, such as weekly sales data, is lower when the information is displayed in bar chart format as opposed to a line graph or point graph:
A new study suggests that the format in which graphs are presented may be biasing people into being too optimistic or pessimistic about the trends in data that the graphs display.
Academics from City, University of London and University College London found that when people who were not experts about a set of data made predictions about how a trend in the data would develop over time, they made lower judgements when the trend was presented as a 'bar chart' type graph as compared to when exactly the same data was presented as a line graph or a graph consisting of a set of data points only.
Nevertheless, across many different types of trend participants consistently thought sales would be lower when the data were presented as bar charts than line graphs or point graphs.
The researchers wondered whether the reason was that in bar charts the area inside the bar is usually heavily shaded and hence visually draws attention to itself, lowering participants' estimates as compared to the other types of graph where there is no shading to attract the eye and attention.
However, in a third experiment, they found the same lower forecasts for bars even when the bars were left unshaded.
In a fourth experiment they tested a version of a bar chart where the bars emanated from the top of the graph rather than the bottom. While subtle trends in the data suggest this may reverse the bias, the findings were inconclusive.
[...] As well as affecting the decisions that individuals make, these biases may also affect the many businesses that perform analyses like 'demand forecasting,' where historical data is used to estimate and predict customers' future demand for a product or service; specifically when these judgements are made unaided by individuals directly 'eyeballing' graphs and estimating how they think a trend is going to develop.
Journal Reference:
Stian Reimers and Nigel Harvey, Bars, lines and points: The effect of graph format on judgmental forecasting, Int. J. Forecasting, 2023. DOI: https://doi.org/10.1016/j.ijforecast.2022.11.003
(Score: 2) by Ingar on Saturday March 11, @11:56AM (3 children)
Playing with scales, offsets, shapes etc.. Marketeers turned this knowledge into science decades ago.
Decisions should be based on the numbers.
(Score: 5, Insightful) by janrinok on Saturday March 11, @01:12PM
Graphs and other visual devices make numbers more easily understood for many people. The important part is to use the appropriate graph for the values you want to display. I think that we all know that it is possible to abuse graphs to mislead people.
But misleading users is not what this story is about. This story is about how different types of graph are either more or less understood even when they are being honest in every respect about the information that they are displaying.
(Score: 0) by Anonymous Coward on Saturday March 11, @01:13PM (1 child)
Agreed. And the best way to make sense of these numbers is to make a graph . . .
(Score: 2) by looorg on Saturday March 11, @08:44PM
While graphs are great and all I still prefer having the numbers instead, that way if needed I can make my own graph or chart or whatever visual display of said numbers I chose. Sometimes a properly formatted table of data is superior to a graph or chart. But, for me, in some regard data >> visual representation of data. That said sometimes you just can't go wrong with a graph/chart of some kind, it's all you need and it does the trick. You don't really care enough to make that effort and in that case the easiest representation of a large set of data is some kind of visual representation beyond the numbers.
(Score: 4, Insightful) by gznork26 on Saturday March 11, @01:56PM (3 children)
Is it just me, or does using a line graph offer a less analytical path to making the projection?
If I'm presented with a line graph and am asked to project where it continues to, the unthinking default is to simply continue on the same slope as the final segment of the line, which I then have to override with the effect of earlier data points. Given a set of raw data such as points or bars, I first have to imagine the connecting lines before making a projection, so analysis happens earlier in the process. In the case of the points, I could also choose to defocus a bit and experience the data as a cloud, which makes it less processing to imagine a mean path through them.
(Score: 2) by Reziac on Sunday March 12, @02:28AM (2 children)
My question is.. which one influenced people toward a more =accurate= prediction? because more optimistic relative to the other type of chart may or may not be more correct.
And there is no Alkibiades to come back and save us from ourselves.
(Score: 3, Interesting) by gznork26 on Sunday March 12, @05:56PM (1 child)
In my case, simply continuing the most recent slope is rarely the most accurate prediction.
(Score: 2) by Reziac on Sunday March 12, @06:17PM
Or why I don't trust truncated data... exactly what slope is now off the chart, and is the visible slope a trend or an anomaly?
And there is no Alkibiades to come back and save us from ourselves.
(Score: 5, Insightful) by istartedi on Saturday March 11, @06:23PM (1 child)
There must be 50 ways to deceive the governed. Just pick from this stack, Jack. Use a short span, Stan...
In all seriousness, this might be more subtle than my two "favorite" ways to present data to suit your bias: 1. The chosen time frame fallacy, in which you zoom in on some part of the data (usually the most recent data) to confirm your bias. and 2. The derivative vs. integral, in which you trumpet the fact that a rate of increase has reached a high level while ignoring the fact that the data is volatile and is likely to have rates of change that vary widely.
Appended to the end of comments you post. Max: 120 chars.
(Score: 2) by krishnoid on Saturday March 11, @11:49PM
It's the period you choose, Cruz [youtu.be], to your point 1.