5.1- Tables and Figures

Welcome to Unit 5 of Writing in the Sciences. This week, I’m going to walk you through how to write a scientific manuscript. Before we jump into the scientific manuscript, however, I do want to briefly continue our discussion from last week on the writing process. I want to give you one more tip on avoiding procrastination and easing anxiety around writing. First of all, writing is hard for everyone. I get intimidated and anxious when I have a big story to write, especially when I haven’t gotten started yet. One thing I do to ease my anxiety is to break the writing up and set very reasonable goals for each day. As I’m driving into work, I might tell myself, my goal today is just to write 400 words. If I can make 400 words, I will have had a successful day. I choose a goal that I know is easy, it’s so reasonable and doable that it doesn’t seem daunting. It doesn’t make me feel like crawling under my desk and curling up in a ball. This makes it easier to get started on the writing that day, and what usually happens is, I reach my goal in a few hours, and by then, I’m on a roll. I feel on track because I’ve met my goal, and I’m able to keep going and get a lot more done that day. But it’s comforting for me to know in the back of my mind that if, for some reason, the writing ends up being really tough that day, and I can only get to 400 words, I can still call that a success. If you set goals too high, like forcing yourself to try to whip out the entire manuscript in one day. This can be so daunting and unrealistic, that it makes it hard to even get started, and you’ll end up procrastinating and get nothing done, so set small, doable goals. Now on to scientific manuscripts, I recommend that you write them in this order. I believe that you need to nail your tables and figures down first. Don’t just put together some shell tables or something rough, but come up with a beautiful, polished set of tables and figures. The tables and figures contain the story of your paper. Each table and figure should have a clear point, and together, they should tell the story of your manuscript. So the first thing that you need to do is to figure out that story that your tables and figures are going to tell. And finalize those tables and figures, make them look professional. Once you have a complete set of tables and figures, the results section falls right out of these, so you should write the results next. The results section just gives a high level summary of each table and figure, so it’s easy to write results if you’ve done the tables and figures well. Next, I suggest writing the methods, you could actually write this at any time, because it’s just a play-by-play of what you did, but I think it makes most sense to write it third. Then, you should write the introduction section. I think it’s important to nail down the story of your manuscript in the tables, figures, and results first, before you write the introduction, because that helps you to know how to frame the introduction. Next, you should write the discussion section, that’s, of course, the hardest to write, because it involves the most writing, it’s the most complex. But having written the rest of the manuscript at this point, it makes it easier to know what to put in the discussion. Finally, always do your abstract last, abstract means to pull out. You are pulling out bits and pieces of the other sections, so it’s much easier to write after you’ve already done the other sections. So this is the order I recommend writing manuscripts in, and I’m going to go over the sections in this order in my lectures for this week. If you want to do some further reading, I’ve given you two references on the page here, and I’m going to be citing these two sources throughout the modules this week. First, there’s a great series in Clinical Chemistry. It goes through all the parts of the scientific manuscript and you don’t need to be a chemist to benefit from these, really these articles apply to any scientific discipline. It’s a very well done and easy to read series. If you want to longer, more comprehensive reference, I recommend Mimi Zeiger’s textbook, the Essentials of Writing Biomedical Research Papers. That’s a comprehensive textbook with lots of practice exercises, it’s especially useful for those in biomedicine. In the first module for this unit, we’re going to talk about tables and figures. In a lot of ways, the tables and figures are the most important part of the manuscript. One big mistake that scientists make is they do the tables and figures kind of haphazardly. They just kind of thow the data into the tables without much thought. You need to think through those tables and figures very carefully, because they form the foundation of your story. Editors, reviewers, and readers may look at the tables and figures first. When I get a paper to review, I skim the abstract just to get a sense of what the paper’s about, and then I jump right to the tables and figures. I want to look at the data for myself before I read the author’s take on their data. This means that the tables and figures have to be able to stand alone, they have to be self-contained. The reader should not need to refer back to the text to make sense of them. Acronyms have to be defined, experimental details need to be defined. Also, each figure and table has to tell a clear story, and they should progress in that story from one to the next. Each table and figure should make a clear point, and as you’re creating the tables and figures, you should know what that point is, and you should try to stick to that point as much possible. Some people would go so far as to say that the tables and figures are the crux of the manuscript, and the writing around them is just window dressing. I picked out a quote here that’s specifically about computational science, but just substitute the word data for the word software here. He says, an article about computational science in a scientific publication isn’t the scholarship itself, it’s merely advertising of the scholarship. The actual scholarship is the complete software development environment and the complete set of instructions which generated the figure. In the same vein, you could say that the data in your tables and figures is the scholarship, is the story, and the writing around the tables and figures is just advertising for them. A couple of tips on tables and figures, first of all, you want to use the fewest figures and tables needed to tell the story. Just as in writing prose, you want to be concise and to the point. Plus, scientific journals may have limits on the numbers of tables and figures that you’re allowed to submit. Second, don’t repeat the same data in both a table and a figure, choose to present that data as one or the other. One of the decisions you’re going to have to make is, which data belong in a table and which data belong in a figure. What’s the difference? Figures should have a visual impact. So if you have data that lends itself to a nice, visually appealing graphic, think about putting it in a figure. Figures are used to show trends and patterns, figures also tell a quick story. So if there’s a single big point that you want to get across to your readers, that’s another thing to put in a figure. I like figures because they can give more information about the data. So if you want to show all of your data points, rather than just giving summary statistics, for example. Or if you want to show the distribution or range of the data, a figure works well for that. People also use figures to highlight the most important result of their paper because, again, Figures are visually appealing, and they can grab the reader’s attention. So if there’s a specific result you want to emphasize, consider putting that in a figure. Tables are used to give precise values. In a figure, you don’t usually convey the precise value down to the decimal place. So if you think those precise values are important, you need to put them in a table. And of course, if you have many values with many variables that you need to display, a table is better for displaying a lot of information. Now let’s talk about the format of tables. Every table has a title, which should identify the specific topic or point of a table. Remember, every table tells a story, and the title should give a quick synopsis of that story. Be sure to repeat the same key terms in the table title, the column headings of the table, and in the main text of the paper. Be consistent, so that you don’t confuse your reader. Remember, it’s okay to repeat keyword you want to repeat them, so that your reader doesn’t get confused. You also want to keep the table title concise, an example title would be descriptive characteristics of the two treatment groups, and then means, plus or minus, standard deviation or N percent. This tells the reader the point of the table to describe the two groups, and let’s them know what statistics to expect. Tables also have footnotes, and different journals have different rules for what symbols to use for footnotes. Some journals might use stars, whereas others might use letters, A, B, C. Don’t try to guess. When you are starting to build your tables and figures, identify the journal that you plan to submit the paper to. And then pick up a published article from that journal, and copy the footnote symbols used in the example. Each journal has its own set of guidelines, and the best way to figure out those guidelines is either to read the author’s instructions that the journal puts out, or even easier, just pull a paper from that journal and copy the format in their published papers. If you’re using footnotes to indicate statistically significant differences, make sure to be clear about exactly what comparisons were made, which groups were compared, and what statistical test was used. Also use the footnotes to explain experimental details or acronyms. Again, this table has to stand on its own. So any acronyms or definitions, or measurement conventions have to be defined within the table. The reader should not have to go back to the text to look those up. In formatting tables, there are a ton of little decision points you have to make, like, do I capitalize the variable names in the table? Do I flush them left or center? Do I use italics? Where do I put the footnotes? What are the footnote symbols? Don’t guess, don’t try to reinvent the wheel and make all this up from scratch. Just go pull a published paper from the journal you are targeting, and copy what you see there. Every journal has its own set of style guidelines. You might as well make your tables look like other published papers in the journal you’re targeting. Reviewers and editors are impressed, when they see professional looking tables in the journals style. Another thing to keep in mind is that the convention for most journals, is that tables contain just three horizontal lines. One above the column headings, one below the columns heading, and one below the data. A lot of people are unaware of this, so they put in many more lines than this, and it ends up looking unprofessional. Here’s an example table with some made up data. When my daughter was two, the Wizard of Oz was her favorite movie. And there’s a line in that movie where the good witch asks Dorothy, Are you a bad witch or a good witch? So they have Bad Witches and Good Witches in Oz. So I made up some pretend data on Bad Witches and Good Witches. In my hypothetical study, I have 13 Bad Witches, and 12 Good Witches. I’ve measured their descriptive characteristics, and put it into a descriptive table, which is your typical table one for a medical study. This is what a typical table looks like. Notice the three horizontal lines. One above the column headings, one below the column headings, and one below the data before the footnotes. That’s how most journals format their tables. Notice that this table has a simple clear point to compare the two groups of witches. And we can quickly glean what’s different about the two groups. Bad witches are older, less healthy, and somewhat less employed. Some journals will also gray out every other row to help readers distinguish one row from the next. Notice there are still just those three horizontal lines, but the shading helps guide the readers eyes. This also looks very professional. Now I’m going to show you some things that you should avoid doing in your tables. Okay, first of all, I mentioned earlier the importance of getting grid lines correct. You should have just those three horizontal lines. Here’s a table in which they have included all of the grid lines. And it might seem like a minor point, but the problem is that the table looks unprofessional. This is not how tables appear in journals. I make a big point of this with my students at Stanford, because I’ve reviewed many, many papers over the years. And at some point I noticed that papers with tables like this, with all the grid lines, tend to have lots of problems elsewhere in the paper. Maybe using all the grid lines like this is more common among new authors, who don’t have much paper-writing experience, or maybe it’s just a mark of general sloppiness. But I noticed this correlation between the presence of the grid lines and problems elsewhere in the paper. So when I flip the tables to the tables and I see all those grid lines, there’s a red flag that goes off in my head. I automatically expect that the paper is going to have other problems. I suspect that other reviewers and editors have formed the same association in their heads, either consciously or unconsciously. And I’ve actually had several students come back to me after taking my course and say, that they started noticing this association between the presences of all the grid lines and poor quality papers. So you don’t want to make this mistake, because it’s a red flag for reviewers and editors. It might seem like a trivial thing, but it signals to reviewers and editors that your paper is sloppy and unprofessional. Always take the time to get those grid lines right. Figure out in your word processing program how to get rid of the extra grid lines. Make your tables look professional. Another thing that doesn’t look professional is, if your columns and data are misaligned. You can see here that items in the columns are not lined up nicely. They don’t usually look as bad as this. I’ve exaggerated for effect here. But you’ll often see numbers that aren’t perfectly lined up, especially when decimal places are involved. I also have some words here beginning with capitals, and others beginning with lower case, I mixed it up. This inconsistency, also looks unprofessional. This kind of sloppiness, makes a big negative impression on reviewers and editors, so take the time to make your tables look professional. Another problem I see with a lot of tables, is people love to go out to a ridiculous number of significant figures. I think that’s because if you put something like age into your statistical analysis program, your computer will spit out values, like the average age or the average BMI, to several decimal places. And then people just cut and past the value into their tables, without thinking about it. But we don’t need to know BMI to the third decimal place, and it just clutter’s the table. Plus, we may not have measured BMI this. Precisely, you should only go out to as many decimals places as you can claim as significant figures. So if you only measured BMI to one decimal place, your summary statistics for BMI should not have more than one decimal place. I generally recommend going out to no more than one decimal place for clinical variables like age, BMI, and exercise. So watch that number of significant figures. Also, always give units for the variables in your table. A lot of my students forget to include units when listing variables like this. I think it’s easy to forget the units when, again, you’re just cutting and pasting output from your statistical analysis program. But the units are often ambiguous. For example, what does exercise mean here? Is that 30 minutes a day? Is that 30 minutes a week? Is that 60 minutes a year? Is that 60 hours a year? There’s no way to tell without having the units. Even age may seem obvious. You may assume that 45 and 36 is referring to years here, but we don’t know for sure. It could be months, if we are talking about toddlers. So units are essential. Another mistake people make on tables is to include too many columns. I had a nice, crisp table to begin with. It was easy to see that the goal of the table was to compare bad witches and good witches. It was easy to glean that bad witches seemed to be older and less healthy than good witches. Adding extra columns just detracts from this nice simple message. Here we have an overall column. That overall column is extra information that doesn’t have to do with the comparison. And if a reader really wanted to know the overall value for some reason, they could calculate it from the other data in the table. So it just adds clutter. Similarly, statistical significance can be neatly indicated using subscripts and footnotes, as I did in the original table. The point of the p-values here is just to flag statistically significant differences, which can be done with a star and a footnote. There may be instances when it’s important to show the exact p-values for all the different variables. But I don’t think that’s the case here. Here, readers just need to be able to quickly glean which variables differ significantly. Extra columns make the table hard to read and detract from the main focus and take-home message of the table. So get rid of them. All right, now moving on to figures. There are three types of figures that we’re going to talk about. First, there’s what we call primary evidence. That would be things like gels, photographs, x-rays, micrographs, pathology slides. They’re in there to show the quality of the data. Also because there’s a seeing is believing element to these. If readers can see the lines on the gel for themselves, they’re more confident in the results. Primary evidence is usually pretty straightforward to create, because it’s just pictures or movies. Second, there’s graphs. These are graphical ways of displaying your data. Things like line graphs, bar graphs, and scatter plots. I’ll talk more about these in a minute. Finally, there are drawings and diagrams. I don’t think people use these enough, actually. Drawings and diagrams are a bit underused in the scientific literature. They have really nice uses. You can use them to do things like illustrate an experimental setup or a workflow. Or to indicate the flow of participants in your study. Or to illustrate cause and effect relationships or cycles. Drawings and diagrams can convey key information to the reader that might be quite tedious and convoluted when written in prose. And sometimes it’s nice you can give a hypothetical model. Or it’s nice to represent things that you can’t see, like microscopic particles or microorganisms. Sometimes it’s helpful to represent those as cartoons. That can be nice for the reader to have a visual of something that they couldn’t otherwise see. So I’m going to go through each of these types of figures now in turn. Of course, all your figures will have a legend. You need the legend so that the figure can stand alone. It should have a brief, informative title that conveys the main point of the figure. It should contain essential experimental details so that the reader doesn’t have to go back and read the main text. You’ll also have a legend to define symbols, lines, or patterns, or to explain what’s in different panels, if you have like an A, B, C, D and E panel. Or you’re going to need also to give statistical detail such as what tests were used and where did those p-values come from, what the p-values were. Here’s an example of a figure legend. This is from a study that found that tomato and Arabidopsis plants kind of eat microbes such E coli bacteria. I’m not going to read through the legend. But you can see that it has an informative title and sufficient experimental details that that figure can stand on its own. This is the figure that goes with the legend I just showed you. You’ll notice there’s a lot of different panels in this figure. They also have some letters and arrows in the panels. You’ll notice all the little white arrows and there are some letters like E, M, and R. All of this has to be explained in the legend. This was a paper where they were showing that plants actually eat microbes, like E coli. So that’s a cool paper with some very cool images. Those green parts are the E coli being taken up by the plant roots. The plants are actually eating the microbes. So this is primary evidence. It’s showing you, hey, here it is. Here is the E coli in the plant roots. Seeing is believing. Here’s another example of primary evidence, a gel. You need to see those bands in the gel to believe the results. So that’s primary evidence. Next I’m going to jump to graphs. I could teach a whole course on graphs. You really need a whole course on data visualization to do justice to this topic. And it’s such an important topic that I’d actually recommend that you do take a course in data visualization in the future. But I’m going to give you the five-minute version. Just to let you know the kinds of graphs that are out there. Among the most common types of graphs are line graphs, scatter plots, bar graphs, individual-value bar graphs, histograms, box plots, and survival curves. I’m going to go through each of these. So line graphs show trends over time or trends over something that’s increasing or decreasing, like age or dose. Typically, we just graph the mean values for each group at each time point. But if the study is very small, you could even show the trend lines for individual people or individual animals. Just to give you an example, here’s a line graph that compares two groups. A treatment group, which is in green, and a controlled group, which is in red. They are tracking what happens to the number of positive cells as you increase the dose of DPI. This graph tells a quick story. Increasing DPI, that’s your x axis variable, doesn’t seem to affect the control group much. The red line’s pretty flat. But you can see it causes the treatment group to dip and then rebound. So it tells a quick visual story. The one thing that’s a bit confusing in this graph is I don’t like the stars on top. These probably have to do with statistical significance. But it’s not immediately obvious what they mean. Okay, bar graphs. Bar graphs are used to compare groups at one time point or one dose. Everybody loves bar graphs because they are easy to understand. They tell a quick visual story, and your reader doesn’t have to do too much work to understand a bar graph. So here are some examples. This was from that study looking at tomato plants eating e coli. They have two controls and a treatment. And the treatment group clearly takes up more e coli. They incorporate more of the microbes. Now, that’s easy to glean from the graph. There are a few things I’m going to point out on this graph that I would improve. For example, it’s not immediately clear to me what the a, b, and c mean. If possible, it’s best to not make the reader have to go to the legend to figure that out, make it something more obvious. It’s not also not terribly beautiful visually because they’ve left the bars with white spaces. Usually, shading those bars makes it a little bit nicer. So here’s another example, this one’s a little bit visually appealing. Notice they’ve shaded the bars in gray. This graph also appears to be at higher resolution. It’s sharper and crisper than the graph I just showed you. I also love the fact that they’ve put the ends, the sample sizes, on each bar to indicate how many are in each group. Normally, people leave this information off. But it’s so important for the reader to have have that information at their fingertips that I love to see it put on the graph like this. So this is a nice bar graph that has a clear take on point. The outcome, averaged degree is increasing over increasing concentrations. Next, scatter plots. Scatter plots are used to show relationships between two variables, usually to continuous or to numeric variables. I love scatter plots because they show all the data. Most graphs and tables just show summary statistics. But you don’t get to see all the individual data points. The scatter plot is great because it shows everything about the data. It shows all the individual data points so it gives you more information. It really gives the reader a sense of what’s going on in the data, instead of just giving one perspective on the data. In a way, it shows all the dirty laundry. So I like scatter plots because you can see those individual data points. Let me give you some examples. This scatter plot is from a study that was looking at high grade and low grade brain tumors. The high grade tumors are in red, the low grade tumors are in blue. And the graph shows the correlation between expression levels of a particular gene that’s on the x-axis, and levels of the biomarker CD44 that’s on the y-axis. You can see that you’re getting a lot more information from a plot like this. The high grade tumors in red have a higher level of both the gene expression and the CD44 biomarker than the low grade tumors. You can see the red is more in the upper right-hand corner than the blue is. But what you can see here that you wouldn’t be able to glean from summary statistics, is that there’s a lot of variation in that red group. Some of the people in the red or the tumors in the red group have very high levels of both, but we also have tumors that have low levels of both. So there’s high variation here. You’ll also notice that the relationship between the gene expression and the biomarker, it’s not a straight line relationship, it’s a bit curved. So you can see that the shape here is not exactly a total linear relationship. So I love scatter plots because they give you so much information. Again, you’re kind of airing your dirty laundry but that allows the reader to get a really good feel for what’s going on in your data. One caution on scatter plots though. If you are going to superimpose either straight lines or curved lines like they have here, just be aware that these lines can fool the eye. In the plot I’m showing here, there is a clear straight-line relationship between x and y. So it’s perfectly appropriate to superimpose a straight line, a regression line, on that plot. But be careful because sometimes lines can fool the eye. So I’m going to show you that now. So here’s a scatter plot that I made. I just made up some data, an x variable and a y variable. And then, I superimposed a line on that. If you just glance at this quickly, it looks like there’s a nice inverse relationship between x and y, right? It’s actually an illusion, though. X and y are completely unrelated. But I superimposed a line slanting down 90 degrees. And it draws your eye in such a way that it makes you see an inverse relationship between x and y. It makes it appear that those two have an inverse relationship. And just to show you that this is really just a visual trick, I also superimposed a 90 degree line in the positive direction. And notice that this is the exact same scatter plot. But when I put the line in the other direction, all of a sudden, your eye starts to think that there is a positive correlation between x and y. But of course, both of these are just your eye being fooled. Because if I take the lines off of this graph, you can see that there is actually no relationship whatsoever between x and y. So it’s easy to. To fool your eye. Be careful when you’re superimposing lines because it can make relationships look stronger than they really are, and this can be misleading. A few tips for graphs, again, graphs are supposed to tell a quick visual story. So you want to keep it simple, you want to make it easy on your reader. Instead of representing different groups with arbitrary symbols such as filled in circles, or unfilled circles, or squares, or triangles, try to be a bit more clever than this. For example, if you have a treatment group and a control group, label the graph with t’s for treatment and c’s for control. This is much easier on the reader because they don’t have to keep referring back to the legend. Also, use different colors for different groups. When I learned to make graphics, we used to have to use dash lines and solid lines to distinguish groups. Because you had to assume that your paper was going to be read in black and white. People would copy papers out of journals, and the copies would always be black and white. Well, now we’re in an era when you can use color. Most everyone is reading papers on an electronic device so you can go ahead and use colors to distinguish different groups. That’s much easier on the reader. Keep in mind that if you’re figure turns out too complex, maybe you should consider putting that data in a table instead. So, let me give you some examples. Here’s an example, the line graphs on the left side of the slide are quite nice. I like those line graphs. You can see that there are four groups and the proportion of cooperation, that’s your y variable, is going down over time in all the groups except one. That group is red seems to be different than the other three groups in terms of the trajectory. So that line graph tells a nice story. But then the authors also added a bar graph to the figure on the right-hand side. And this is summarizing the same data that are in the line graph. What they did is they averaged sessions five through ten. So that they could show that the red group was, on average in those five sessions, statistically higher than the other three groups. The stars are supposed to be indicating statistically significant differences here. But it’s very confusing. Those stars, it’s not clear what comparisons they’re referring to. Plus, seeing the data points from the individual sessions is much more informative. Informative than just seeing an average of an arbitrary number of sessions. So I don’t think you gain anything by adding the bar graph to this figure. It just serves to clutter up the graph and confuse the reader. The authors could have just added a star somewhere on the line graph to indicate that the red group differed from the other three based on the statistical test that compares the trajectories over time. One other tip here, notice that the line graph uses squares, triangles, diamonds, and circles to differentiate the different groups. It’s a little hard on the reader because the reader has to keep looking back to the legend to figure out which group is which. Maybe the authors could have labeled the groups, the group names on the graph somewhere or used more informative plotting symbols. Here’s another example where the graph has too much going on. It’s not telling a clear and simple story. The bar graphs are too cluttered, there’s too much data. There are no clear patterns, so it’s not obvious what the take home point is supposed to be. If the data are this complicated, you maybe better off presenting them in a table rather than a figure. A figure is supposed to tell a simple visual story and I don’t think this one accomplishes that. This figure is from the same paper, has the same problems. Finally, I’m going to jump now into diagrams and drawings. You have diagrams and drawings at your disposal. And these are under used, I think in the scientific literature. They can be very powerful. For example, many things don’t lend themselves well to text. Such as explaining the flow of participants in a study or the details of an experimental setup. Often a picture is better for conveying this information. Also if you have some kind of hypothetical cause and effect model that you’re trying to propose. A diagram is great for indicating how you think those variables interact, especially if the relationships are complex. This may all be better put in a picture than trying to explain it in some complicated text. Also if you’re talking about something people can’t see like virus or proteins or antibiotics, a cartoon can help your readers visualize what you’re talking about. So think about in your paper, if there’s places where you could use a diagram or drawing to your advantage. I’ll give you some examples. So here’s a figure from a paper that was looking at drug company advertisements and journal. This drawing sums up the typical layout of an advertisement. So you see that at the beginning, at the top of the advertisement page, you get the picture of the happy physician or happy patient. And then you get at the bottom right-hand corner, you always have your Kaplan-Meier survival curve. A picture is worth 1,000 words here, if you tried to describe this set up in prose, it would be uninteresting to read. It would be boring to read and it would be much harder to convey that information. This is visual information, so might as well put it in a drawing. I love this diagram. This was in a paper describing dog bites and cat bites in human patients. Rather than just sticking the data into a boring table, the authors were clever enough to put it in a picture. The key messages are instantly clear. Most bites happen on the hands, and there’s some difference in the location of cat bites versus dog bites. That’s a much more interesting and nice way to present the data than just putting it in a boring table. Diagrams are great for drawing causal diagrams like this. You can see how this is much simpler to present in a diagram than to try to explain all those relationships prose. And finally, here’s a cartoon illustration of the creation of some novel antibodies. Again, the picture is really helpful to the reader in understanding how these antibodies are built. And the final thing I want to mention is that nowadays, you’re not just limited to tables and figures. Your papers can have videos, this is back to the paper on plants eating microbes for nutrients? Seeing is believing applies even more for movies than for still pictures. So we can play this little video and actually see the microbes being taken up by the plant.