|
Post by format on Apr 4, 2013 14:39:59 GMT -5
Hello,
Do you have any advice on formatting long tables for articles you will submit to be reviewed?
I feel like a few categorical independent variables, say income or education, and my table is more than one page.
A reviewer just complained about my table being 3 pages long but I could see no other way to have it be legible. I'm fine with having insignificant covariates mentioned in a footnote but not for reviews; I want reviewers to see them.
Thoughts?
Thanks, Format
|
|
|
Post by putitintheirhands on Apr 4, 2013 15:05:33 GMT -5
A reviewer who has a problem with your table length should suggest a solution. If the reviewer complained without suggesting a solution, they are not doing their job.
So, I would toss it back in their lap: "Although I'm sympathetic to Reviewer A's concerns, I feel that leaving all coefficients in the table is necessary to preserve the integrity of the results. However, if the reviewer will suggest alternatives (including what could be omitted), I will consider his/her suggestions."
|
|
|
Post by yes on Apr 4, 2013 17:37:40 GMT -5
The above is decent advice, but I would put it on the editor:
"... I am happy to consider alternatives, such as removing insignificant variables, if the editors would like."
|
|
|
Post by socyuser on Apr 4, 2013 18:17:31 GMT -5
I'm not sure it's completely the reviewer's responsibility, I spend a hell of a lot of time figuring out how to make my figures and tables publishable. As a reviewer I might get annoyed at a person who didn't make the same effort. Since you say you would be fine with a shortened version of the table in the final published piece, I would make a version of the shortened table for the submission and then also have the full version of the table at the end of the submitted manuscript with a note saying that the full table is at the end for the purpose of the review. My suggestion for shortening: if you are including some variables simply as controls, I often just group them into a line that says "controls included" and then have yes or no in the columns for the different models rather than the coefficients, then a note in the text saying that the full tables are available upon request.
|
|
|
Post by format on Apr 4, 2013 19:37:23 GMT -5
Dear Socyuser,
Question: what do you do about categorical controls? Lets say, for example, that you measure education as four categories. Do you list all four with yes/no? do you just list education and if so, how do you determine whether it's yes/no.
Thank you for your sage counsel, format
|
|
|
Post by highflyer on Apr 4, 2013 23:28:31 GMT -5
I'm not sure it's completely the reviewer's responsibility, I spend a hell of a lot of time figuring out how to make my figures and tables publishable. As a reviewer I might get annoyed at a person who didn't make the same effort. Since you say you would be fine with a shortened version of the table in the final published piece, I would make a version of the shortened table for the submission and then also have the full version of the table at the end of the submitted manuscript with a note saying that the full table is at the end for the purpose of the review. My suggestion for shortening: if you are including some variables simply as controls, I often just group them into a line that says "controls included" and then have yes or no in the columns for the different models rather than the coefficients, then a note in the text saying that the full tables are available upon request. This ^ is reason number 341 why sociology is a mess. A reviewer should do their job. Their job is to evaluate the merits of the research. Not the placement of the legend in the figures. Not table design. Not the placement of commas, or the preferences for saying "Odds-ratios" vs. "odd-ratios" (though clearly the former is preferable--it's just, with all they have to do, why address THAT?). It is the editor's job to tell the author to cut the tables if the editor wants that. If the author had done as the poster above says, the reviewer could still have gotten exasperated. "Why do you have two copies with different information for the same table?" as they chomp down their lunch, surf the web, and skim your paper at the same time. Please don't say it can't happen, I've seen it. Or, worse, the reviewer just gets angry and takes it out on other aspects of the paper that are perfectly fine--"Don't estimate one model with interactions by gender for everything, estimate one model for women and one for men!" The former is better because you can test coefficients, but try to tell some of the people in the field this. Or, try to tell some people in the field that such minutia is not what reviewing is about. A good editor would inform the author to either ignore reviewer drivel, or exactly what the editor wants in that regard (cut the insignificant coefficients? cut the controls, significant or not? something else?). Alas, the field has few good editors. So, basically, they send you the reviews and say, "Here ya go." Perhaps it is because the people who seek editorships these days are more likely to be administrator types rather than intellectual types. So it becomes a paper processing exercise, not the nuanced, critical, essential task of sifting the chaff from the grain, and nurturing good works into the discussion. For the administrator types, a reviewer saying the tables are too long is really important. To the intellectuals, its a sign the reviewer is an idiot, and probably the rest of their review is laced with equally inappropriate claims, so their claims would be discounted. But, the intellectuals have other interests so they don't seek editorships for the most part. So the field heads further into disaster, as the administrators pat each other on the back for a job well done--yay, our acceptance rate is now 10%, down from 14% just two years ago. YAY! We reject so many papers, we must be REALLY GOOD! DOUBLE YAY!
|
|
|
Post by socyuser on Apr 4, 2013 23:34:34 GMT -5
Well, first of all I should say that I'm assuming that this table is some sort of regression-like table. Also, I do this more for control variables that are mainly in the model for the sake of controlling for them and where the actual coefficients are not important parts of my paper. I'm not sure if this will work with formatting but I'll try to give an example. (Argh. Ok, I just previewed it and it looks like all of my efforts at formatting are completely ruined, but perhaps you can puzzle it out anyway. If not, you can send me a message through this system and I can try to send you a formatted version). Suppose my table looks like this:
Model 1 Model 2 Model 3 Model4 Key ind variable 1.42* 1.20* 1.11* 1.01*
Female 0.5* 0.5* 0.3* Age 0.01* 0.02* 0.0 Race (white omitted) Black -0.6* -0.4* -0.2 Asian 0.1 0.1 0.0 Other -0.3 -0.2 -0.1
Education (less than HS omitted) HS 0.3 0.1 Some college 0.7* 0.4* College 1.1* 0.8*
Major occupation (professional omitted) Manager -0.1 Clerical -0.4 Service -1.1* Manual -1.6*
In other words, in model 1 I'm establishing a key relationship of interest. In model 2 I'm adding in several demographic controls (female, age, race groups), but these controls are not key aspects of my study. They do have an impact on the slope for my key independent variable so I will probably discuss these results in the text, and maybe even discuss which of the demographic controls had the most effect, but since these controls are not a central part of this paper showing the coefficients for each of these variables is not crucial. On the other hand, perhaps a key hypothesis is that education will have a big effect on the question I am studying, which I'm testing in model 3. In this case showing the coefficients for the education categories is probably important. Finally, model 4 adds categorical variables for occupation group, but maybe here the coefficients for each of the occupation groups are not a critical. So, I might compress the results like this:
Model 1 Model 2 Model 3 Model4 Key ind variable 1.42* 1.20* 1.11* 1.01*
Demographic controls included? yes yes yes
Education (less than HS omitted) HS 0.3 0.1 Some college 0.7* 0.4* College 1.1* 0.8*
Occupation included? yes
In other words, think about why you are including the different variables in the model, whether they are all equally important for the narrative of your paper, and whether the coefficients for each individual category is important to show. I prefer to think this way rather than just removing all of the non-significant variables simply because they aren't significant (if they are part of a key hypothesis, the fact that they aren't significant is itself important).
I realize, though, that this approach may not fit your approach to your analysis. The example I'm using here, which is common in my research, is establishing a basic relationship and then seeing if it "goes away" after adding in a variety of controls. In this approach the coefficients for all of the controls may be less important than the effect of adding the controls on the slope for the key independent variable. In this case, I think collapsing groups of variables or categories of a variable works well.
If instead your paper is about examining a particular outcome and looking for which of a long list of variables is correlated with an outcome (rather than looking at the effects of adding controls), then perhaps the coefficients for each value is more important and this approach won't work. Hope this is useful.
|
|
|
Post by socyuser on Apr 4, 2013 23:45:43 GMT -5
I agree with highflyer that editors should do more in providing guidance on these things, but I also don't think that this means that authors can just submit unrefined stuff and expect reviewers and editors to tell them how to make it publishable. I think everyone agrees that authors are responsible for crafting a well-argued and well-written paper. Well, an important part of that narrative is readable and organized tables and figures.
|
|
|
Post by pre on Apr 5, 2013 8:45:23 GMT -5
Here's the tables that socyuser was trying to present. (The markup tag is "pre", as in "preformatted.") I'll reproduce the entire post.
Well, first of all I should say that I'm assuming that this table is some sort of regression-like table. Also, I do this more for control variables that are mainly in the model for the sake of controlling for them and where the actual coefficients are not important parts of my paper. I'm not sure if this will work with formatting but I'll try to give an example.
Model 1 Model 2 Model 3 Model 4
Key ind variable 1.42* 1.20* 1.11* 1.01*
Female 0.5* 0.5* 0.3* Age 0.01* 0.02* 0.0 Race (white omitted) Black -0.6* -0.4* -0.2 Asian 0.1 0.1 0.0 Other -0.3 -0.2 -0.1
Education (less than HS omitted) HS 0.3 0.1 Some college 0.7* 0.4* College 1.1* 0.8*
Major occupation (professional omitted) Manager -0.1 Clerical -0.4 Service -1.1* Manual -1.6*
In other words, in model 1 I'm establishing a key relationship of interest. In model 2 I'm adding in several demographic controls (female, age, race groups), but these controls are not key aspects of my study. They do have an impact on the slope for my key independent variable so I will probably discuss these results in the text, and maybe even discuss which of the demographic controls had the most effect, but since these controls are not a central part of this paper showing the coefficients for each of these variables is not crucial. On the other hand, perhaps a key hypothesis is that education will have a big effect on the question I am studying, which I'm testing in model 3. In this case showing the coefficients for the education categories is probably important. Finally, model 4 adds categorical variables for occupation group, but maybe here the coefficients for each of the occupation groups are not a critical. So, I might compress the results like this: Model 1 Model 2 Model 3 Model 4
Key ind variable 1.42* 1.20* 1.11* 1.01*
Demographic controls included? yes yes yes
Education (less than HS omitted) HS 0.3 0.1 Some college 0.7* 0.4* College 1.1* 0.8*
Occupation included? yes
In other words, think about why you are including the different variables in the model, whether they are all equally important for the narrative of your paper, and whether the coefficients for each individual category is important to show. I prefer to think this way rather than just removing all of the non-significant variables simply because they aren't significant (if they are part of a key hypothesis, the fact that they aren't significant is itself important). I realize, though, that this approach may not fit your approach to your analysis. The example I'm using here, which is common in my research, is establishing a basic relationship and then seeing if it "goes away" after adding in a variety of controls. In this approach the coefficients for all of the controls may be less important than the effect of adding the controls on the slope for the key independent variable. In this case, I think collapsing groups of variables or categories of a variable works well. If instead your paper is about examining a particular outcome and looking for which of a long list of variables is correlated with an outcome (rather than looking at the effects of adding controls), then perhaps the coefficients for each value is more important and this approach won't work. Hope this is useful.[/quote]
|
|
agree with reviewer
Guest
|
Post by agree with reviewer on Apr 5, 2013 8:50:13 GMT -5
It is your job to present your data is a sensible and compelling way, both in your writing and in your tables. Take a look at examples published in the journal. A three-page table seems ridiculous. Learn how to edit it to present it professionally. The reviewer is not your copy editor.
|
|
|
Post by Excel on Apr 5, 2013 9:32:46 GMT -5
You should be able to type all of that in an excel sheet, copy it, then paste it in word as a special image. No way that takes up three pages.
|
|
|
Post by highflyer on Apr 5, 2013 9:35:28 GMT -5
It is your job to present your data is a sensible and compelling way, both in your writing and in your tables. Take a look at examples published in the journal. A three-page table seems ridiculous. Learn how to edit it to present it professionally. The reviewer is not your copy editor. Nobody asked the reviewer to be the copy editor. Why do reviewers volunteer? Just address the scholarly merits of the content. Let the author have their voice. Please.
|
|
|
Post by format on Apr 5, 2013 12:04:12 GMT -5
Thank you all for your thorough answers.
I think the big difference between what is presented here and what I do is that I include the standard errors. I've been using the outreg2 package in Stata.
I thought it was expected to include the errors; am I wrong?
Thanks again
|
|
|
Post by socyuser on Apr 5, 2013 13:02:44 GMT -5
Thank you all for your thorough answers. I think the big difference between what is presented here and what I do is that I include the standard errors. I've been using the outreg2 package in Stata. I thought it was expected to include the errors; am I wrong? Thanks again Yes, do include the standard errors in the table. I just didn't include them in the example to make the task of writing the post quicker. I have left out standard errors in powerpoint presentations, but for published papers they are necessary. So, unfortunately eliminating standard errors is not the solution. One possible workaround is if your tables are long but not very wide (only one or two models) is that I have seen some papers where the standard errors are in a separate column rather than underneath.
|
|
|
Post by socyuser on Apr 5, 2013 13:08:12 GMT -5
Here's the tables that socyuser was trying to present. (The markup tag is "pre", as in "preformatted.") I'll reproduce the entire post. [/quote] Thanks!
|
|