Descriptive Statistics Tables—Which Variables to Include? What Summary Statistics to Report? Raw or Transformed Variables?

1. Introduction: Table 1 Is the First Table Your Readers See—But You May Never Have Thought Carefully About It

In your empirical paper, the first table is usually the "Descriptive Statistics" (Summary Statistics) table. You include the dependent variable, the key independent variable, a few control variables, run a summarize command, and paste the output into a table. You may think this is merely a "formatting requirement"—nothing substantive, reviewers will only glance at it, and it doesn't require much thought.

Then you receive the reviewer comments:

"The minimum value of variable X in Table 1 is −3.2, but this variable should not be negative in theory—please explain."

"The standard deviation of the dependent variable is 15 times its mean—is the data severely right-skewed? If so, has the author considered using a logarithmic or other transformation?"

"Table 1 presents 12 variables, but variable Y (mediator) and variable Z (moderator) are not included—please add them."

"The sample size in the descriptive statistics table is N = 4,567, but the sample size in the regression table is N = 3,021—the difference of more than 1,500 observations between the two needs to be explained."

You are stunned—Table 1 is not just a formatting requirement. It is a "data health check report," the first checkpoint for reviewers and readers to judge whether your data are reasonable and your analysis credible.

Core message: A descriptive statistics table conveys the most critical information about your data in the least space. Its purpose is not to "list some means and standard deviations"—but rather (1) to let readers understand the basic characteristics of your sample and the distributional range of your variables, (2) to help you and reviewers identify potential problems in the data (outliers, coding errors, skewed distributions, sample selection, etc.), (3) to provide ex ante factual justification for your model specification (e.g., whether to take logs or winsorize), and (4) to document the fundamentals of the estimation sample used in your subsequent regressions. Regarding which variables to include: the dependent variable and key independent variable must be included, and all control variables used in subsequent regressions should also be included—reviewers need the descriptive statistics to assess whether your control variables have sufficient variation to identify effects and whether there are signs of collinearity. Mediators and moderators—if used in your main analysis, they should appear in the descriptive statistics; if only used in auxiliary analyses, they can be placed in the appendix. Regarding which summary statistics to report: the number of observations N, mean, standard deviation, minimum, and maximum are the minimal necessary set. For skewed variables, add the median or p25/p75. Regarding raw versus transformed variables: report the version you actually use in your regressions—if you use ln(income) in your regressions, the descriptive statistics table should report ln(income), not raw income. You may additionally report statistics for the raw variable in the appendix. Regarding the questions reviewers most often raise: inconsistent N across variables, implausible minimum/maximum values, standard deviations far exceeding means (severe right skew), binary variable means near 0 or 1 (lack of variation), extreme outliers persisting after winsorization, and large discrepancies between the descriptive statistics sample and the regression sample—these are the six key points reviewers check in Table 1.

2. Which Variables Should a Descriptive Statistics Table Include?

2.1 Must Include: Dependent Variable + Key Independent Variable + All Control Variables

Why must these three tiers of variables appear in the descriptive statistics table:

• Dependent Variable: Readers need to know the mean, range, and distribution of your Y—this is the baseline for understanding the "magnitude" of your results. If the mean of your dependent variable is 0.08 and your regression coefficient is 0.02, readers know that the effect of X is approximately 25% of the mean of Y.
• Key Independent Variable: Reviewers need to assess whether your key X has sufficient variation (whether the standard deviation is large enough, whether the range is wide enough) and whether there are anomalous extreme values (whether the max is 10,000 times the min).
• Control Variables: Reviewers need to assess whether your control variables are "real variables"—if a control variable has a standard deviation near zero (almost no variation), it is effectively a constant and should not be in the regression. Also, if a control variable's mean or range is very close to that of the key X → possible collinearity.

Principle: If you use a variable in your regressions—it should appear in the descriptive statistics table. There should be no "invisible" additions or deletions of variables between the descriptive statistics and the regressions.

2.2 Mediators and Moderators—Include or Not?

Mediator: If the mediator is part of the "causal mechanism from X to Y"—in the standard "X → M → Y" structure, M is central to the analysis. M should appear in the descriptive statistics table because it helps readers understand "the distribution and variation of this mechanism variable in the sample"—which directly affects your interpretation of the mediation effect.

Moderator: If the interaction term is used in your main analysis—the moderator itself should appear in the descriptive statistics. The interaction term ($X \times W$) itself typically does not appear in the descriptive statistics table (it is a constructed quantity without independent descriptive meaning), but W itself needs to appear.

Practical advice:

• If the mediation/moderation analysis is central to your paper (e.g., "the purpose of this paper is to study the mechanism through which X affects Y via M") → M and W appear in the main descriptive statistics table.
• If mediation/moderation is only a robustness check or supplementary analysis → M and W appear in the appendix and may be omitted from the main table.

2.3 Variables That Do Not Need to Appear in the Descriptive Statistics Table

• Fixed effects dummies (unit FE, year FE, industry FE): No need to report the means of 5,000 individual dummies.
• Interaction terms: Not needed—their statistics can be derived from their components.
• Instrumental variables: If it is an "exclusion-restriction instrument" (e.g., quarter of birth, policy shock), it should appear in the descriptive statistics table—because reviewers need to see the variation and distribution of the IV. If it is a "technical IV" (e.g., lagged values, industry-mean IV), it can be presented in a dedicated table in the "IV Validity" section of the paper.

2.4 A Practical Checklist

Does your Table 1 include:
□ All dependent variables used in the regressions (including Y for different specifications)?
□ The key independent variable(s) (all key X's)?
□ All control variables?
□ The mediator (if used in the main analysis)?
□ The moderator (if used in the main analysis)?
□ The instrumental variable (if IV is used)?
□ Sample restriction variables (e.g., screening variables for "urban sample only" or "A-share listed firms only")?

3. What Summary Statistics to Report?—The Minimal Necessary Set and Extended Set

3.1 Minimal Necessary Set—N, Mean, Standard Deviation, Minimum, Maximum

These are the five statistics every descriptive statistics table must include—reviewers use them for the first "health check" of your data:

* Stata one-liner to output these five statistics
summarize y x1 x2 x3

3.2 Extended Set—Median, p25, p75, Skewness

For the following types of variables, the minimal necessary set is insufficient—you need to add additional statistics:

When to add the median (p50):

• When the mean and median differ substantially (e.g., mean = 50, median = 8) → the data are severely right-skewed, with the mean pulled up by extreme values.
• When reviewers notice that SD >> Mean—the median shows the true center of the "typical individual," unaffected by extreme values.

When to add p25 and p75:

• When you use subgroup analysis (e.g., "high-education group vs. low-education group")—showing the p25 and p75 of years of education helps readers understand the meaning of your group cutoffs.
• When you dichotomize a continuous variable—showing p25 and p75 lets readers see how variation changes before and after grouping.

summarize y x1 x2, detail   // includes p25, p50, p75, skewness, kurtosis

3.3 Group-Specific Descriptive Statistics—"The Grouped Version of Table 1"

In difference-in-differences (DID) analyses, randomized controlled trials (RCTs), or any paper with a "treatment group vs. control group," the descriptive statistics should present statistics by group—with a "difference test" (t-test or standardized difference) at the right end.

                Treatment Group      Control Group       Difference
                Mean (SD)            Mean (SD)           p-value / Std. Diff
DV              0.34 (0.18)          0.31 (0.17)         0.04
X               12.3 (4.1)           12.1 (4.0)          0.52
...

Purpose: To show whether the treatment and control groups are "balanced" on observable characteristics before treatment. If the two groups differ substantially on pre-treatment characteristics → your parallel trends or conditional independence assumption for DID or matching is already questionable on observable dimensions.

3.4 Descriptive Statistics for Panel Data—"Overall, Between, Within"

For panel data, a more refined approach decomposes the sources of variation:

                Overall SD      Between SD      Within SD
DV              0.35            0.28            0.19
X               12.3            10.8            5.2

* Panel variation decomposition
xtsum y x1 x2

Purpose: To show whether the variation in X comes mainly from between-individual or within-individual differences—which is crucial for understanding the identification source of fixed effects models (which use only within variation). If the within SD is extremely small → identification in the FE model may be weak.

4. Raw or Transformed Variables?

4.1 Basic Principle: Report the Variables You Actually Use in Your Regressions

If your regressions use the logarithmic form—the descriptive statistics table should report the logarithmic form:

* If ln_wage is used in the regressions
gen ln_wage = ln(wage)
* Report ln_wage in the descriptive statistics, not wage

Reason: Readers need to see the distribution of "the variables that enter the regression model"—not the distribution of a raw-scale variable that is not directly used by the model. If you use $\ln(\text{income})$ in your regressions but Table 1 reports income—readers cannot infer the distributional characteristics of $\ln(\text{income})$ from the mean and standard deviation of income.

4.2 When Can Both Columns Be Reported?

In some papers, the descriptive statistics report both the raw and transformed variables—for example:

Variable          N      Raw Mean (SD)        Transformed Mean (SD)
Income            4500   85000 (120000)       11.2 (0.9) [ln(income)]

Applicable scenarios:

• Your readership is accustomed to understanding variables in their original units (e.g., "income of 85,000 yuan").
• The interpretation of coefficients after log transformation requires returning to the original scale ($\beta$ in a log-log model is an elasticity—readers need to compare it with the raw mean).

4.3 Winsorized Variables—Report the Winsorized Version

If you winsorize, the descriptive statistics table should report the winsorized variables—because these are the versions entering the regressions. You can add a note in the descriptive statistics table or state it in the "Data" section of the paper:

"All continuous variables are winsorized at the 1st and 99th percentiles. Table 1 reports descriptive statistics for the winsorized variables."

* Report after winsorization
winsor2 x, replace cuts(1 99)
summarize x

Do not mix pre-winsorization and post-winsorization values in the same Table 1—readers will be confused about which version was used in the regressions.

4.4 Standardized Variables—Report the Standardized Version

If your regressions use standardized variables (z-scores), the descriptive statistics will have means ≈ 0 and SD ≈ 1—readers may not intuitively understand "the effect of one standard deviation." In such papers, some authors choose to report descriptive statistics for both the standardized and raw versions—or at least provide the mean and standard deviation of the raw variable (in a footnote or appendix) so readers can translate the standardized results back to the original scale.

5. The Purpose of a Descriptive Statistics Table—Before Entering Regressions, What Does Your Data Say?

5.1 Purpose One: Let Readers See Your Sample—"Who Are You Studying?"

Descriptive statistics give readers a "snapshot of the sample"—without reading the full methods and data sections, readers can quickly learn from Table 1:

• The average years of education in your sample is 9.2 → your sample may be concentrated among low-education groups.
• The mean of your dependent variable (whether one starts a business) is 0.07 → only 7% start a business; this is a rare event—your subsequent Logit regression may face rare-event bias.
• In your firm sample, the mean of R&D expenditure is 0 and the median is 0 → more than half of firms have no R&D—this is a corner solution piled up at zero.

This information does not automatically surface in regressions—but reviewers need it to understand your results.

5.2 Purpose Two: Expose Data Problems—"Did You Do Your Homework Before Running Regressions?"

This is the most underappreciated function of descriptive statistics: it is an ex ante diagnostic table for data quality.

Before running regressions, you should already have used these signals in Table 1 to determine whether the data need cleaning, transformation, or adjustment:

• Are Min and Max within theoretically reasonable ranges?
• Is the standard deviation far larger than the mean?
• Are binary variable means close to 0 or 1?
• Is N consistent across variables?

If these problems are only discovered after the regressions have been run—reviewers will conclude that you "ran regressions without checking the data"—which is more serious than admitting "the data have limitations."

5.3 Purpose Three: Provide Factual Basis for Subsequent Model Choices

"Why did you take the log of this variable?"—Because SD >> Mean in Table 1 (severe right skew), and log transformation is the standard treatment.

"Why did you winsorize?"—Because Max / Mean > 100 in Table 1, indicating extreme outliers.

Table 1 is the first half of the "ex ante factual basis" for your model specification—you discovered problems in the descriptive statistics and addressed them in the model specification. This forms a logical closed loop before reviewers: problem → discovery → treatment → robust results.

6. What Problems Require Checking Your Data?—Seven Red Alerts in Descriptive Statistics

6.1 Alert One: N Differs Significantly Across Variables

If, in Table 1, some variables have much smaller N than others (e.g., N drops from 5,000 to 2,000 to 800) → missing values are proliferating. You need to check:

• Is the missingness random? Or are certain types of individuals more likely to have missing values?
• If your final regression sample is N = 800 while the full sample for the key variable is N = 5,000 → reviewers will ask you to explain why 84% of observations were dropped and whether this sample selection affects the external validity of your conclusions.
• Presenting a "sample construction process" in the paper—each step of attrition from the raw data to the estimation sample with reasons—is the standard way to address this concern.

6.2 Alert Two: Minimum or Maximum Values Are Theoretically Implausible

• Age with Min = −3 or Max = 210 → coding errors or outliers in the data.
• Firm size with Min = 0 → every firm has at least one employee; zero is impossible.
• Income with Min = −50,000 (in pre-tax data) → could be self-employment losses, which need to be clearly flagged and confirmed as reasonable.

When reviewers see implausible min or max values, they will ask you to explain or correct them—check yourself before submission.

6.3 Alert Three: Standard Deviation >> Mean (Severe Right Skew)

SD > 3 × Mean → the data are severely right-skewed—a few extremely large values inflate the standard deviation. Consider:

• Log transformation (ln(y) or ln(y+1)).
• Winsorization.
• Or using median regression as a robustness check.

6.4 Alert Four: Binary Variable Means Near the Boundary

• Dependent variable (whether exporting) mean = 0.02 → only 2% of firms export. Your Logit model faces an "extremely low event rate" problem—this needs discussion (rare-event bias, Firth Logit, or using LPM).
• Dependent variable mean = 0.98 → only 2% of firms do not export—the same problem.

6.5 Alert Five: Extremely Small Standard Deviation (Lack of Variation)

• A control variable with SD = 0.01 and mean = 3.5 → this variable barely varies in the sample. Why is it included as a control? Its coefficient cannot be reliably estimated (collinearity or lack of variation).

6.6 Alert Six: Key Variable Means Differ Substantially Before and After Winsorization

• If your descriptive statistics table reports winsorized variables, but the pre-winsorization Max = 50,000 (post-winsorization Max = 120) → the raw data contain very extreme outliers. Is the winsorization threshold reasonable (1st and 99th percentiles? Or 0.5th and 99.5th?)—reviewers may ask.

6.7 Alert Seven: Difference Tests in Group-Specific Descriptive Statistics Are Largely Significant

• If, across 15 pre-treatment characteristics, 12 have difference-test p-values < 0.05 between your treatment and control groups → your treatment is not "as-if random"—the two groups already differ on observable characteristics, and unobservable characteristics are also likely to differ. This is a major red flag for DID or matching analyses.

7. Six Types of Reviewer Criticisms Most Commonly Directed at Table 1

7.1 Criticism One: "The Descriptive Statistics Table Is Missing Key Variables"

"Table 1 presents the key independent variable and control variables, but the mediator M, although used in the subsequent mediation analysis, does not appear in the descriptive statistics. Please add it."

Response: Add M and W to Table 1 (if they are part of the main analysis) or to the appendix (if they are robustness checks).

7.2 Criticism Two: "The Sample Size in the Descriptive Statistics Table Differs from That in the Regression Table—the Gap Is Large and Unexplained"

"Table 1 has N = 8,234, but Table 2 (baseline regression) has N = 4,567. The difference of more than 3,600 observations between the two needs to be traced and explained."

Response: Construct a "sample selection flowchart" in the paper—from the raw data to the estimation sample, step by step: (1) raw observations → (2) drop observations with missing key variables → (3) drop observations failing sample restriction criteria → (4) final estimation sample. Report N and the reason for each step. Reviewers should not have to guess why your sample changes—they need you to present the complete chain.

7.3 Criticism Three: "Minimum/Maximum Values Are Theoretically Implausible"

"The minimum of firm age is −5. Please explain why firms appear in the sample before their founding year."

Response: Check the raw data—it could be a coding error (incorrect year format leading to negative age calculations), a data entry error, or a special case in certain industries (e.g., how firm age is calculated after mergers). Whatever the cause—it needs to be corrected or explained. If it cannot be corrected—frankly flag this anomaly in the paper and show whether conclusions change after excluding these anomalous observations.

7.4 Criticism Four: "The Key Variable Is Severely Right-Skewed but Not Transformed—Why?"

"The standard deviation of the key independent variable R&D expenditure is 20 times its mean—the data are severely right-skewed. Why did the author choose to model in levels rather than on the log scale?"

Response: If there are reasons to model in levels (e.g., model specification required by theory, coefficients in levels have a specific economic interpretation, or log-scale results are similar in robustness checks), state them explicitly in the paper. If there are no such reasons—consider switching to a log specification, or at least showing robustness on the log scale.

7.5 Criticism Five: "The Binary Variable Mean Is Extremely Low—The Event Rate Is Too Low to Support Nonlinear Models"

"The mean of the dependent variable (whether a patent is granted) is 0.008. At this event rate, the Logit model may suffer from rare-event bias. Has the author considered using Firth Logit or a linear probability model as alternatives?"

Response: In this situation, reviewers typically expect you to at least discuss the possibility of rare-event bias—even if your main model remains unchanged, showing in robustness checks that conclusions are robust to Firth Logit or LPM usually resolves this criticism.

7.6 Criticism Six: "The Descriptive Statistics Table Reports Raw Variables, but the Regressions Use Transformed Variables"

"In Table 1, the mean of income is 85,000, but the variable used in the regressions is ln(income). Readers cannot learn the distribution of ln(income) from Table 1—Table 1 should report the version entering the regressions."

Response: Replace income in Table 1 with ln(income), or report both. This is extremely easy to fix—but it is also repeatedly raised by reviewers because it signals a disconnect between the author's data preparation and regression specification.

8. Common Misconceptions

8.1 Misconception One: "Descriptive Statistics Is Just Pasting the summarize Output"

Descriptive statistics are not a mechanical operation—they are a "health check report." Before pasting the summarize output into a table, you should read it yourself first—check whether any of N, min, max, mean, or sd makes you frown. If you do not read it yourself—reviewers will read it for you, and their conclusion is usually "the author does not seem to have carefully checked the data."

8.2 Misconception Two: "N Differing Across Variables Is Fine—Stata Automatically Handles Missing Values"

Yes, Stata automatically handles missing values—but reviewers need to know why N differs, how large the differences are, and whether the differences could affect the results. If the N differences are small (e.g., 5,000 vs. 4,800) → a footnote may suffice. If the N differences are large (e.g., 5,000 vs. 2,000 vs. 800) → an explanation in the main text is required.

8.3 Misconception Three: "The Descriptive Statistics Table Should Be as Concise as Possible—Include Fewer Variables and Statistics"

Conciseness is good—but "omitting key information" is not conciseness; it is omission. A good descriptive statistics table is complete in information—and refined in presentation. Convey the most critical information in the least space. Additional less important details can go in the appendix.

8.4 Misconception Four: "Constructed Variables (Such as Interaction Terms and Squared Terms) Need to Appear in the Descriptive Statistics"

Interaction terms and squared terms do not convey independent information—they are constructed from their components. Readers do not need to know the mean and standard deviation of X²—these can be derived from the mean and standard deviation of X. The descriptive statistics table should report the base variables before construction.

8.5 Misconception Five: "Descriptive Statistics Can Be Written After the Regressions—It Is Only a Formatting Requirement Anyway"

Descriptive statistics should be the first substantive operation in your empirical analysis—before you run your first regression. By examining the descriptive statistics, you will discover data problems, determine whether logs or winsorization are needed, and assess whether binary variables have sufficient variation. If you only look at the descriptive statistics after all regressions are run—you may have run dozens of regressions on problematic data and need to redo the work.

9. Summary

Seven Core Takeaways for Descriptive Statistics Tables:

1. Variables that must be included: dependent variable, key independent variable, and all control variables used in the regressions. Mediators, moderators, and instrumental variables—if used in the main analysis → include them; if used in the appendix → include them there. Interaction terms and fixed effects dummies → not needed.

2. Minimal necessary statistics: N, mean, standard deviation, minimum, maximum. For skewed variables → add the median and p25/p75. For panel data → use xtsum to show between and within variation.

3. Report the version entering the regressions—if your regressions use ln(y), Table 1 reports ln(y), not y. If the raw version needs to be shown → put it in the appendix.

4. Three purposes of descriptive statistics: (a) let readers see your sample, (b) expose data problems (before regressions), and (c) provide ex ante factual basis for model specification.

5. Seven data red alerts: inconsistent N, implausible min/max, SD >> Mean, binary means near boundaries, extremely small SD (lack of variation), large gaps before and after winsorization, and largely significant group difference tests.

6. Six classic reviewer criticisms: missing key variables, unexplained N inconsistency, theoretically impossible min/max, severe skewness without transformation, rare events, and variable version mismatch. All six are preventable in advance—before submission, put yourself in the reviewer's shoes and read through your Table 1.

7. Table 1 should be your first operation, not your last. Do descriptive statistics before running regressions—use them to check your data quality—and only then proceed to regression analysis. If the order is reversed, your work may need to be redone at the review stage.

One-sentence closing:

"Table 1 is the 'facade' of your paper—but its function is not decorative. It is the health check report of your data: the mean is your data's temperature, the standard deviation is your data's breathing, and min and max are your data's blood pressure—each has a normal range. Inconsistent N across variables is telling you 'someone has fallen behind—who are they?'. SD >> Mean is telling you 'there are extreme values in the long tail—can I tolerate them?'. A binary variable mean near 0 is telling you 'the event is too rare—can my model handle it?'. A good descriptive statistics table starts asking questions before you run your regressions. A bad descriptive statistics table is just a string of numbers you pasted from Stata to LaTeX, waiting for reviewers to read it for you—and then waiting for them to ask the questions. You are not 'submitting a formatting assignment'—you are telling reviewers and readers 'I know my data, I know its boundaries and limitations, and I entered my regression analysis with that knowledge.'"

10. Presentation Suggestions for Bilibili/WeChat Official Account

• Bilibili video: Use the "health check report" as the core visual metaphor. Opening: a doctor (researcher) holding a health check report (Table 1), facing a patient (dataset). The doctor points at the numbers on the report: "Temperature (mean) is normal. Heart rate (standard deviation) is a bit fast—your variable is right-skewed. Blood pressure min/max—the lowest value is outside the normal range; there is a coding error in your data that needs investigation." Voiceover: "Table 1 is not a formatting assignment—it is the health check report of your data. Every statistic is a vital sign of your data. Mean, standard deviation, min, max—they answer not 'what is the variable called'—but 'is this data healthy?'. Good researchers read the health check report before running their first regression. Bad researchers first discover the problems when reviewers read it for them." Scene One "Six classic reviewer criticisms": six envelopes are opened one by one—each contains a reviewer's question: ① "Why is the mediator not in Table 1?" ② "Why did N drop by 3,000 observations between Table 1 and Table 2?" ③ "Why is the minimum income −500?" ④ "SD is 20 times the Mean—why not take logs?" ⑤ "Binary dependent variable mean 0.008—is Logit reliable under such rare events?" ⑥ "Table 1 reports income, but the regression uses ln(income)—why the inconsistency?" Voiceover: "Six types of criticisms—all of which you can discover and fix yourself before submission. Reviewers are not your proofreaders—they expect you to have done the first step of data work in Table 1." Scene Two "What to include and what not to include": the screen shows a Table 1 template—the left column is "Must Include" (dependent variable, key X, control variables—green), the middle is "Optional" (mediator, moderator—light green—depending on whether they are in the main analysis), and the right is "Not Needed" (dummies, interaction terms, squared terms—light red—excluded). Annotation: "Table 1 should present—the base variables used in the regressions that have independent economic and statistical meaning. Not all variables need to appear in this first table."

• WeChat Official Account: Create an infographic of the minimal necessary statistics (N, Mean, SD, Min, Max) paired with "what each statistic means in the eyes of reviewers." Turn the seven data red alerts into health signal cards—each alert paired with an action recommendation for "what to check." Turn the six classic reviewer criticisms into "prevention cards"—check your Table 1 against this card before submission. Turn the variable tier decision table (must include, optional, not needed) into a three-tier decision reference card—especially useful for beginners. Turn the raw vs. transformed variable placement rules into a comparison card. Turn the grouped descriptive statistics template (treatment vs. control + difference test) into a template card.

• Recommended titles:

  • Main title: "Descriptive Statistics Tables—The First Health Check Report for Your Data"
  • Alternative title: "What Variables Should Table 1 Include and What Statistics Should It Report?—The Minimal Necessary Knowledge for Descriptive Statistics"
  • New media title: "What Are Reviewers Looking at When They Read Your Table 1?—Six Problems You Can Prevent Before Submission"

• Key quotes:

  "Table 1 is not the output pasted from Stata's summarize command—it is the first step in getting to know your data. The mean tells you where the 'center' of your sample is, the standard deviation tells you how wide the 'breathing range' of your variable is in the sample, and min and max tell you where the boundaries of your data are. Reviewers will look at these four numbers—you should look at them first."

  "SD >> Mean—your data are gasping in the long tail. When reviewers see this, they will ask why you did not take logs. A binary mean near 0 or 1—your binary variable barely varies. Reviewers will ask whether your nonlinear model is still reliable when events are too rare. These are not questions to think about after running all your regressions—you should start thinking about them at Table 1."

  "The descriptive statistics table is the first table readers see in your paper—and the first place reviewers check. Inconsistent N, implausible min/max, and variable versions mismatched with the regressions—these are problems reviewers can spot in the third second. If you spend three minutes discovering and fixing them in advance—you can save three rounds of review."

  "The variables in Table 1 should be the variables you actually use in your regressions. If the regression uses ln(income) but Table 1 reports income—reviewers' first reaction: 'The author may not have noticed—or the data preparation and regression analysis are disconnected.' This kind of criticism is entirely avoidable—align the variable names, forms, and definitions between Table 1 and the regressions. Align, and you will not be asked."