About this example
This page is the results chapter of an example doctoral thesis in corporate finance. The study asks which firm-level and country-level factors are associated with leverage in non-financial companies listed on the Dubai Financial Market (DFM). It uses secondary data on 12 firms over five financial years (2008 to 2012), giving 60 firm-year observations. The chapter shows how to report descriptive statistics, correlations and a regression model for a small panel honestly: one variable per table, the right outcome and predictors, standard errors that respect the repeated observations, and conclusions that match the evidence.
It is an illustrative teaching example, not a real student's work, and no firm is named. All numbers in the tables are illustrative data, created and checked for internal consistency for teaching; they are not DFM statistics and must not be quoted as findings about UAE companies. Read each passage of the example, then the "What to notice" note beneath it. Do not submit any part of it as your own work; your university's academic integrity rules apply.
The example: data and descriptive statistics
4.1 Introduction
This chapter reports the results of the analysis described in Chapter 3. Section 4.2 describes the sample and variables. Section 4.3 presents descriptive statistics, Section 4.4 the correlations among the variables and Section 4.5 the regression of leverage on its candidate determinants. Interpretation in the light of trade-off, pecking-order and agency explanations (Myers, 1984; Jensen & Meckling, 1976; Harris & Raviv, 1991) is left to Chapter 5.
4.2 Sample and variables
The sample comprises 12 non-financial companies listed on the DFM with complete annual reports for 2008 to 2012, giving 60 firm-years. Leverage (LEV), the outcome, is total debt divided by total assets. Following the determinants most often examined in cross-country work (Rajan & Zingales, 1995), the firm-level predictors are firm size (SIZE, the natural logarithm of total assets in AED), profitability (ROA, net income divided by total assets) and growth (GROWTH, the annual change in revenue). Two country-level variables, real GDP growth (GDP) and consumer price inflation (INF), take one value per year and are therefore the same for every firm in a given year.
4.3 Descriptive statistics
| Variable | Mean | SD | Min | Max | Skewness | Excess kurtosis |
|---|---|---|---|---|---|---|
| LEV (debt / total assets) | 0.29 | 0.09 | 0.03 | 0.45 | −0.61 | 0.03 |
| SIZE (ln total assets) | 20.96 | 0.72 | 19.79 | 22.75 | 0.50 | −0.15 |
| ROA | 0.042 | 0.033 | −0.038 | 0.106 | −0.34 | −0.29 |
| GROWTH (revenue change) | 0.05 | 0.15 | −0.38 | 0.41 | −0.33 | 0.63 |
| GDP growth (%) | 1.9 | 3.5 | −4.5 | 5.0 | −1.06 | −0.36 |
| Inflation (%) | 3.2 | 4.4 | 0.7 | 12.0 | 1.49 | 0.23 |
Mean leverage was 0.29 (SD 0.09): on average, debt financed just under a third of total assets. Yearly means ranged from 0.27 (2011 and 2012) to 0.31 (2009). Skewness and excess kurtosis for the firm-level variables lie between −1 and 1. Because SPSS and most packages report excess kurtosis, a normal distribution corresponds to a value of 0, not 3. Normality of the raw variables is not an assumption of regression; the relevant checks concern the residuals and are reported in Section 4.5 (Ernst & Albers, 2017).
The example: correlations
4.4 Correlations
| LEV | SIZE | ROA | GROWTH | GDP | |
|---|---|---|---|---|---|
| SIZE | −0.32 (0.013) | ||||
| ROA | −0.33 (0.010) | −0.03 (0.80) | |||
| GROWTH | 0.29 (0.023) | −0.17 (0.21) | 0.15 (0.24) | ||
| GDP | −0.17 (0.20) | 0.04 (0.77) | 0.38 (0.003) | 0.06 (0.66) | |
| INF | 0.05 (0.69) | −0.07 (0.61) | 0.04 (0.76) | 0.06 (0.67) | 0.10 (0.43) |
Leverage was negatively correlated with firm size (r = −0.32, p = 0.013) and with ROA (r = −0.33, p = 0.010), and positively correlated with growth (r = 0.29, p = 0.023). Its correlations with GDP growth (r = −0.17) and inflation (r = 0.05) were small and their confidence intervals included zero. Operating profit scaled by total assets was also considered as a profitability measure, but it correlated r = 0.94 with ROA; the two measure the same construct, so only ROA is used below (variance inflation factor about 8.7 when both are included, against 1.0 to 1.2 for the other predictors). These p values treat the 60 firm-years as independent and are therefore too small; Section 4.5 corrects for this.
The example: regression of leverage on its determinants
4.5 Regression model
Leverage was regressed on SIZE, ROA, GROWTH, GDP and INF by ordinary least squares. Because each firm contributes five observations, the errors for the same firm are unlikely to be independent, and standard errors were clustered by firm (Wooldridge, 2010). With only 12 clusters, confidence intervals use the t distribution with 11 degrees of freedom. Residuals were plotted against fitted values and each predictor to check for curvature and unequal spread (Appendix C), and leaving out one firm at a time did not change the direction of the size, ROA or growth coefficients.
| Predictor | b | Clustered SE | 95% CI | p | p if firm-years treated as independent |
|---|---|---|---|---|---|
| SIZE | −0.034 | 0.019 | −0.076 to 0.009 | 0.11 | 0.019 |
| ROA | −0.98 | 0.54 | −2.16 to 0.20 | 0.09 | 0.004 |
| GROWTH | 0.18 | 0.07 | 0.03 to 0.34 | 0.02 | 0.011 |
| GDP growth | −0.001 | 0.003 | −0.007 to 0.005 | 0.75 | 0.77 |
| Inflation | 0.001 | 0.001 | −0.002 to 0.003 | 0.57 | 0.75 |
Together the predictors accounted for 31% of the variance in leverage (adjusted R² = 0.25). Holding the other variables constant, a one-unit higher revenue growth rate was associated with leverage 0.18 higher (95% CI 0.03 to 0.34). The estimates for size (b = −0.034 per log unit of assets) and profitability (b = −0.98) were negative, in line with the correlations, but their confidence intervals were wide and included zero once the clustering of observations within firms was taken into account. GDP growth and inflation showed no clear association with leverage; with only five distinct values each, the data cannot separate their effects from other events in the same years.
4.6 Summary of results
- Leverage averaged 0.29 of total assets and varied little between years.
- Higher revenue growth was associated with higher leverage (b = 0.18, 95% CI 0.03 to 0.34).
- Size and profitability were negatively associated with leverage, but the estimates were imprecise (95% CIs −0.076 to 0.009 and −2.16 to 0.20).
- No clear association was found for GDP growth or inflation, which the five-year window can test only weakly.
How to adapt this to your own study
- Define the outcome and each predictor once, with units and the formula used, before reporting any number.
- Scale accounting variables (profit by assets, debt by assets) so that they do not simply measure size.
- Report descriptives in one table and comment on the features that matter for the analysis.
- Check predictors that measure the same construct (correlation, VIF) and keep one of them.
- Treat repeated observations of the same firm as a panel: cluster standard errors or use a panel estimator chosen in your methodology chapter (Wooldridge, 2010).
- Report b, standard error, 95% CI and exact p for each predictor, with R² and n in the caption.
- Write the summary as estimates with uncertainty, and keep explanation for the discussion.
Common mistakes in this chapter
- Pasting text from another project (here, a paragraph on satellite elevation data) into a results chapter.
- Copying the interpretation of one table under another table's heading.
- Including the outcome variable, or a transformation of it, among its own predictors.
- Reversing outcome and predictors in the summary ("leverage predicts firm size").
- Treating 60 firm-years as 60 independent observations without clustering or a panel method (Wooldridge, 2010).
- Entering two measures of the same construct (ROA and operating profit) and interpreting both.
- Judging kurtosis against 3 when the software reports excess kurtosis, or testing normality of raw variables instead of residuals (Ernst & Albers, 2017).
- Writing that a hypothesis is "accepted", or that a factor "impacts" the outcome, from an observational association (Harrell, 2015).
| What examiners look for | Where this example does it | Check in your own draft |
|---|---|---|
| Clear variable definitions | 4.2 defines each variable and unit | Can a reader rebuild each variable? |
| Tables that match the text | Tables 4.1 to 4.3, one variable per row | Does every sentence match the table it cites? |
| A correctly specified model | 4.5: outcome only on the left-hand side | Is any predictor a version of the outcome? |
| Inference that respects the design | Clustered SEs, t(11) CIs, naive p shown for contrast | Have you allowed for repeated observations? |
| Estimates with uncertainty | b, SE, 95% CI and p reported | Do you report more than p values? |
| Honest limitations | Five years of macro data stated as a limit | Have you said what your data cannot show? |
| Results kept separate from discussion | Theory left to Chapter 5 | Have you moved interpretation out of the results? |
Key takeaways
- Define every variable, with its unit and formula, before reporting any result.
- Never place the outcome, or a version of it, among its own predictors.
- Repeated observations of the same firm need clustered standard errors or a panel estimator (Wooldridge, 2010).
- Report estimates with confidence intervals and avoid "accepted" or "proved" (Greenland et al., 2016).
- Label teaching numbers as illustrative; in your thesis every number must come from your own data.
