The way your results section should be written up depends heavily on the discipline in which you are working. Refer to a style guide of the professional association in your discipline (e.g., the American Economic Association, the American Sociological Association, American Psychological Association) or to a style guide developed by your institution.
Personally, I like results sections that follow this structure:
Do not copy and paste the results from the output window of your statistical analysis software. They typically contain too much information.
Always start with a bivariate analysis. Table 4 provides an example of how you can report it.
Discuss the results along the lines of your hypotheses, for example: “In line with our hypothesis on gendered giving, women were found to be more likely to give than men. The difference is about 10 percentage points. The negligible difference in empathic state, however, suggests that empathy is not a likely explanation of the gender difference in giving. Neither is the level of education a likely explanation, as the values in the final column show.”
A good table is self-explanatory. Its title, contents, and footnotes allow the reader to understand the design and the results of the analyses without having to read the accompanying text.
Graphs are a good way to present bivariate analyses. When you plan to conduct multiple regression analyses, always inspect a set of scatter plots first, before you proceed to include control variables or test for mediation. You may discover that the distributions of the variables you are interested in are not normal. As an example, consider the plots in Figure 19 from the Datasaurus package (Locke, 2018), all reflecting the same means and correlations.
If the scatter plot looks like one of these, how much confidence can you have in the results of a regression of Y on X? Before you run a model with an interaction to test for moderation, graphically display the relationship you think is moderated by values of the moderator variable. If the graphs do not look too different, the regression results may be misleading.
Start your results section with the most simple description you can think of. For instance, if you are testing the association between religiosity and prosocial behavior, start by presenting simple differences between religious groups, without any covariates. If you report results of an experiment, show the distribution of the outcome variables by condition. Do not jump into analyses of variance and regressions right away, without looking at the distribution of variables.
For an example of how to report regression results, see Table 5 (taken from Bekkers & Schuyt, 2008, p. 90). The table presents odds ratios for the relation between predictor variables and the likelihood of volunteering outside church. Note that the table does not report confidence intervals for the odds ratios, as would be common in other disciplines, such as health research or economics. The table shows three columns. In the first column the denominational differences are displayed, controlling for differences in gender, age, town size, income, and level of education. The logic of the table is that in each successive model, a set of variables is added that tests a pathway of influence from denomination to volunteering. The second column adds predictors from the ‘conviction’ explanation arguing that denominations affect volunteering through social values.
The accompanying text is the following: “Church members also dominate voluntary work in non-religious organizations (see Table 5). Catholics are very well represented in this category of active citizens. Dutch Catholics may not often volunteer for church-related groups and may not be very generous, but they are very active in non-religious voluntary associations. In addition, we see that also older, more highly educated people, people living in smaller communities, and people with lower incomes are more often volunteering in non-religious organizations. The significance of size of municipality is interesting: in research from the U.S. it is often argued that differences between urban and rural areas can be attributed to the different composition of the local population (Wilson 2000; Wuthnow 1998). This does not seem to be the case in the Netherlands. Model 2 reveals intriguing findings: compared to the non-religious, the greater activity of Protestants in volunteer work outside church is due to some of their social values, but the higher volunteer activity of Catholics in non-religious organizations cannot be explained in this way. Altruistic values, and to a smaller extent also generalized social trust, increase the likelihood of volunteering for non-religious organizations. These results partially support hypothesis 4. In contrast to the analysis of non-religious giving, we find no significant relationships of prosocial value orientation, social responsibility or salience of religion on non-religious volunteering. As predicted by hypotheses 2 and 3, model 3 shows that greater exposure to requests for contributions and stronger social pressure to honor these requests promotes voluntary work outside the church. In contrast to hypothesis 1, church attendance, however, actually lowers the likelihood of participation in non-religious voluntary work. Similar findings are reported in U.S. studies (Campbell and Yonish 2003; Park and Smith 2000).”