Is Excel-generated chart acceptable for publication? [closed]

Question

I have always wondered, why it seems I have never seen people using excel-generated charts for publishing. Is excel considered less professional/not serious? Say I have a time-series like this. What could have other chart making platforms do better (that excel cannot) to improve it? The dimension (boundary) of the chart?

Answer 1

Excel can produce high quality charts, but it takes a lot of work. The biggest difference between Excel and other systems is the quality of the defaults. I suspect some people are using excel charts, but they have gone to a lot of effort to tidy them up.

In the example chart, you will, for example, definitely need to deal with the x-axis labels overlapping the y-grid lines. In fact the y grid lines are too strong - in fact I'd put some more consideration into the gird lines in general - why y lines, but no x lines - there might be a good reason, but you should think about it?

Limit the y-axis to make 0 the minimum (I'm assuming a negative flux makes no sense).

I would personally also add a y axis and x-axis border line. You are almost certainly going to want tick marks on the axes. And markers on the lines/larger markers if they are already there but too small.

Your axis titles need to have the same font size (the y looks bigger than the x to me at the moment) and the difference in font size between the labels and the titles is probably too big. I probably take out half of the x-axis labels and make the remaining ones bigger.

One thing that Excel is bad with, and isn't easily fixable is consistency. Getting two plots looking exactly the same, with exactly the same size plot, same size text, with all the elements in the same place is really difficult because it relies so much on the mouse and manually moving/sizing things.

Answer 2

Azor Ahai's comment about reproducibility is the most important one. I used Excel for many years in a professional but non-academic setting but gave it up when I needed auditable, reproducible results. It is extremely easy to make mistakes when you use Excel and very hard to detect that you have made them. It is effectively impossible to check an Excel spreadsheet that someone else has produced and, as we are all taught in basic coding courses, "somebody else" is you in 6 months' time.

If you want other people to be able to follow what you have done and reproduce it, don't use Excel.

Answer 3

You have asked several important questions.

1. Are Excel-generated plots ever accepted in published manuscripts?
2. Is Excel considered less professional/not serious [at creating plots]?
3. What can other chart making software do better (that excel cannot)?
4. What are some examples of things that can be done to improve the sample plot provided here?

In response to your first question, the answer is, unfortunately, yes. Being a regular peer-reviewer for academic publications and an expert in data visualisation, it is usually very easy to identify how graphics have been generated, e.g. whether they have been generated using Excel, ggplot2 in R, MATLAB, or other software. This is because there are hallmark tell-tale signs - usually default settings that aren't changed by the user, which is a dead giveaway. It is worth noting that I have never seen this in Q1 SJR-ranked journals (top quartile of journals), but it does sometimes happen in Q2 and less reputable journals.

In response to your second question, the answer is a resounding "YES". While Microsoft has made large strides to improve their graphical tools (for example, by consulting with data visualisation guru Edward Tufte), the default options are poorly chosen (and even go against Tufte's 5 Principles of Graphical Excellence), are difficult and tedious to manipulate when the default options are untenable, and the amount of work required to create the same plot with different data or for another user to recreate the same plot with the same data is difficult. From the choice of colour schemes to unnecessary details like 3D effects are all bad design choices. Moreover, some types of plots are available which shouldn't be (like pie charts - there is no good reason to ever use pie charts), and many types of plots and features are unavailable (polar charts, interactive plots, GIS maps, 3D surface and 2D contour plots, raster plots, etc.). Compare this to the list of plots available in ggplot2.

As an extension of the response above and in response to your third question, other software excels at making plots (excuse the pun) in many ways:

• The default options are usually better looking (higher quality) and meet Tufte's Principles of Graphical Excellence.
• Changing the default options is usually straightforward.
• Plots generated programmatically (say from R or MATLAB) are able to be reproduced by simply re-running the code (code which can be published alongside a manuscript as supplementary material which facilitates reproducibility).
• Editing the plot at a later date is easier. This is important when you submit a paper for review and reviewers ask you to make changes to your plots. Many features are difficult or impossible to replicate in Excel plots, which leaves you with only one option - taking a screen-capture and manually processing the changes. The tediousness of this task is exacerbated by the number plots to edit and level of detail/changes to make.
• Journals often have strict standards on graphics, such as what format (file extension) is acceptable and often specify a minimum resolution/DPI (usually between 300 DPI and 600 DPI). This is easily accomplished in software like R and MATLAB where the user can specify the desired DPI and re-generate the plot accordingly.

In response to your last question, Ian Sudbery has already provided a number of good points for improving your sample plot. I would add the following:

• Consider using a better colour scheme; the current one is neither grey-scale friendly or colour-blind friendly. I suggest making use of ColorBrewer to help choose good colours for qualitative data, and the Colour-blindness simulator.
• Consider making the lines thicker (I would have to test this to see if it's actually better or not).
• Consider using log-y-scale (if flux is always positive; in which case the error bars should be capped at 0 too) to facilitate comparisons when values are smaller (which is where most of your data occurs).
• The title of the chart is missing (although you may have chosen just not to embed it in the plot itself, which is fine).
• In addition to Ian's suggestion of making the y-axis minimum zero, the maximum should be 550 rather than 650.