Figure 6: Presenting a single regression model using a dotplot with error bars
Figure 7: Using parallel dotplots with error bars to present two regression models
Figure 8: Using “small multiple” plots to present multiple regression models
Table 2 from Stevens (2006): Determinants of Authoritarian Aggression
Variable | Coefficient (Standard Error) |
---|---|
Constant | .41 (.93) |
Countries | |
Argentina | 1.31 (.33)^{### B,M} |
Chile | .93 (.32)^{### B,M} |
Colombia | 1.46 (.32) ^{### B,M} |
Mexico | .07 (.32)^{A,CH,CO,V} |
Venezuela | .96 (.37)^{## B,M} |
Threat | |
Retrospective egocentric economic perceptions | .20 (.13) |
Prospective egocentric economic perceptions | .22 (.12)^{#} |
Retrospective sociotropic economic perceptions | -.21 (.12)^{#} |
Prospective sociotropic economic perceptions | -.32 (.12)^{##} |
Ideological Distance from president | |
Ideology | |
Ideology | .23 (.07) ^{###} |
Individual Differences | |
Age | .00 (.01) |
Female | -.03 (.21) |
Education | .13 (.14) |
Academic Sector | .15 (.29) |
Business Sector | .31 (.25) |
Government Sector | -.10 (.27) |
R^{2} | .15 |
Adjusted R^{2} | .12 |
n | 500 |
^{###}p < .01, ^{##}p < .05, ^{#}p < .10 (two-tailed) | |
^{A} Coefficient is significantly different from Argentina’s at p < .05; | |
^{B} Coefficient is significantly different from Brazil’s at p < .05; | |
^{CH} Coefficient is significantly different from Chile’s at p < .05; | |
^{CO} Coefficient is significantly different from Colombia’s at p < .05; | |
^{M} Coefficient is significantly different from Mexico’s at p < .05; | |
^{V} Coefficient is significantly different from Venzeluela’s at p < .05 |
Our graph
Table 4 from Ansolabehere and Konisky (2006): Registration effects on turnout in New York and Ohio counties: Fixed effects model, 1954–2000
Dependent Variable = County-Level Turnout | ||||||
---|---|---|---|---|---|---|
Full sample (1) | Excluding counties w/partial registration (2) | Full sample w/state-year dummies (3) | Full sample (4) | Excluding counties w/partial registration (5) | Full sample w/state-year dummies (6) | |
% of county with registration | –0.039^{##} (0.003) | –0.036^{##} (0.003) | –0.051^{##} (0.003) | –0.037^{##} (0.003) | –0.034^{##} (0.003) | –0.050^{##} (0.003) |
Law change | –0.020^{##} (0.005) | –0.018^{##} (0.005) | –0.023^{##} (0.006) | |||
Log population | 0.048^{##} (0.011) | 0.036^{##} (0.012) | 0.017 (0.010) | 0.047^{##} (0.011) | –0.035^{##} (0.021) | 0.016 (0.010) |
Log median family income | –0.133^{##} (0.013) | –0.142^{##} (0.014) | 0.050^{##} (0.013) | –0.131^{##} (0.013) | –0.139^{##} (0.014) | –0.049^{##} (0.013) |
% population with h.s. education | 0.071^{#} (0.028) | 0.070# (0.029) | 0.011 (0.024) | 0.072^{#} (0.028) | 0.071^{#} (0.029) | 0.013 (0.024) |
% population African American | –0.795^{##} (0.056) | –0.834^{##} (0.059) | –0.532^{##} (0.044) | –0.783^{##} (0.055) | –0.822^{##} (0.059) | –0.521^{##} (0.044) |
Constant | 1.47^{##} (0.152) | 1.70^{##} (0.171) | 0.775^{##} (0.124) | 1.45^{##} (0.152) | 1.68^{##} (0.170) | 0.819^{##} (0.127) |
R^{2} | 0.91 | 0.91 | 0.94 | 0.91 | 0.91 | 0.94 |
N | 3572 | 3153 | 3572 | 3572 | 3153 | 3572 |
Note. ^{#}p < .05, ^{##}p < .01. Huber-White standard errors in parentheses. Year dummies and state-year dummies are not reported. |
Our Graph
Download the R code for this graph.
To create this graph, you will also need a function that we wrote called plot.reg, available here.
The following graph is identical to Figure 6, except we add vertical tick marks to indicate 90% confidence intervals (i.e. when p < .10).
Download the R code for this graph.
Notes
In general, we don't recommend using multiple confidence intervals, as 95% intervals will usually suffice. (The Journal of Politics, for example, instructs authors not to “use varying numbers of asterisks to indicate different levels of statistical significance.”) However, in cases where multiple intervals are desirable, tick marks (or, alternatively, lines of different width) are useful for differentiating intervals.
We also created functions that exemplify how regression plots could be done automatically. You will need the Zelig R package to try this out. According to their authors, “Zelig is a single, easy-to-use program that can estimate, help interpret, and present the results of a large range of statistical methods.” The functions are in plotReg.R, which also produced Figure 8.
Download the R code for this graph.