Exploration of cover color distributions

---

In this analysis, I examine the disribution of colors in the book covers in our corpus. I analyzed the covers in HSV space. HSV space is a color space that was designed to correspond to more closely align with human perception, (compared to RGB space).

• H = hue
• S = saturation
• V = value (lightness)

Findings:

• Books for girls and older kids have covers that are less saturated (there’s no interaction).
• Girl books have more purple/pink; Boy boys have more red. These patterns are apparent when you look at the covers themselves.
• There aren’t any interpretable trends for advert age?

Saturation and Value

Gender

Larger values = more associated with girls.

measure	estimate	statistic	p.value	conf.low	conf.high
mean_s	-0.1695296	-2.597438	0.0100036	-0.2924741	-0.0410718
mean_v	0.1102376	1.674758	0.0953527	-0.0193975	0.2362274

Advert Age

measure	estimate	statistic	p.value	conf.low	conf.high
mean_s	-0.1492083	-2.3132140	0.0215752	-0.2714768	-0.0222001
mean_v	-0.0351076	-0.5385217	0.5907267	-0.1618140	0.0927376

## 
## Call:
## lm(formula = value ~ earliest_advert_level_months + gender_score, 
##     data = .)
## 
## Residuals:
##       Min        1Q    Median        3Q       Max 
## -0.006743 -0.002169 -0.000170  0.001870  0.008271 
## 
## Coefficients:
##                                Estimate Std. Error t value Pr(>|t|)    
## (Intercept)                   7.132e-03  5.339e-04  13.359  < 2e-16 ***
## earliest_advert_level_months -3.038e-05  1.350e-05  -2.250  0.02541 *  
## gender_score                 -3.146e-03  1.072e-03  -2.934  0.00369 ** 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.00293 on 223 degrees of freedom
##   (16 observations deleted due to missingness)
## Multiple R-squared:  0.05377,    Adjusted R-squared:  0.04529 
## F-statistic: 6.337 on 2 and 223 DF,  p-value: 0.002106

Older books and books for girls are less saturated, but there’s no interaction. Book age and gender score explain half the variance in color saturation.

Hue

Here’s the overall distribution of hue across all book covers.

Predicting color distribution

In these analyses, I tried to predict the proportion of a particular color in book covers as a function of book gender and target age.

Gender rating - facet

This plot shows the correlation between gender rating and proportion color for each of 36 different hues.

Gender rating

This plot is the same data as on the facetted plot, but here a bar is shown for each hue which represents the magnitude of the correlation between the proportion of that color and gender rating.

Advert Age