1. Is there a correlation of touch ratios between convex and concave models?

Null hypothesis: There is no correlation of the touch ratios of different countries between the two different models, as the models are physically different from each other.

Method: using the circular mask for country shapes, only consider touches in the areas of the circles

cor(s$concave,s$convex) 
[1] 0.8974139

The correlation is very high!

Where is the object touched more - at the base or in the countrycircles? About 75% of the touches are in the countrycircles; there is relatively small variation across conditions and model type.

Now let’s look at much much the ring components were touched in each model in each condition. The touch ratios for each ring of the models seem also correlated across both models.

cor(g$concave,g$convex)
[1] 0.7601623

2. How significant are the touch ratio differences between countries and models?

Linear regression of the touch ratio (using the circlemask) with countries as dummy variables with the model type (convex concave) and condition (1-3) as a controls.

The country is Italy, Norway, Spain, and United Kingdom are significantly less touched with Norway, having considerable effects (~8%). Condition 2 is on average 2% less touched.

seg_touch %>% lm(formula = touchratio_country ~ as.factor(convex) + country ) -> m1
summary(m1)

Call:
lm(formula = touchratio_country ~ as.factor(convex) + country, 
    data = .)

Residuals:
     Min       1Q   Median       3Q      Max 
-0.30792 -0.12418 -0.00365  0.10143  0.47819 

Coefficients:
                      Estimate Std. Error t value Pr(>|t|)    
(Intercept)            0.25942    0.02699   9.613  < 2e-16 ***
as.factor(convex)1     0.02482    0.01707   1.454 0.146779    
countryItaly          -0.16114    0.03621  -4.450 1.14e-05 ***
countryNorway         -0.14367    0.03621  -3.968 8.72e-05 ***
countryPoland         -0.00699    0.03621  -0.193 0.847021    
countrySouth Korea    -0.04264    0.03621  -1.178 0.239666    
countrySpain          -0.09502    0.03621  -2.624 0.009043 ** 
countrySwitzerland     0.02368    0.03621   0.654 0.513537    
countryUnited Kingdom -0.13038    0.03621  -3.601 0.000361 ***
countryUnited States  -0.01790    0.03621  -0.494 0.621439    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 0.1659 on 368 degrees of freedom
Multiple R-squared:  0.1427,    Adjusted R-squared:  0.1217 
F-statistic: 6.806 on 9 and 368 DF,  p-value: 4.642e-09

3. To what extent are touch ratios correlated with the shape and position of country circles?

Why is there such a strong correlation btw. convex and concave? must have to do with a combination of: position on board and shape of the country circle. They are all moderately correlated.

cor(touch_model_country_agg$touch_mean,touch_model_country_agg$sd)
[1] 0.3765896
cor(touch_model_country_agg$touch_mean,touch_model_country_agg$area)
[1] 0.4057662
cor(touch_model_country_agg$touch_mean,touch_model_country_agg$elevation)
[1] 0.4140155
cor(touch_model_country_agg$touch_mean,touch_model_country_agg$amplitude)
[1] 0.3236832
#pairs(touch_model_country_agg[c(6,12:16)])

touch_model_country_agg %>% #filter(country!="Switzerland") %>%
  pivot_longer(names_to = "metrics", cols = c(area,elevation,amplitude,sd)) %>%
  ggplot() + aes(x = value, y = touch_mean, color = as.factor(convex)) + geom_point() + facet_wrap(~metrics, scales = 'free')

Modeling based on the strongest correlated model parameter, elevation while controlling for position shows only a tiny insignificant effect, slightly higher if particular parts of the model are considered, e.g. the center.

touch_model_country %>% #filter(country!="Switzerland") %>%
lm(formula = touchratio_country ~ elevation + as.factor(convex) + x*y) %>% summary()

Call:
lm(formula = touchratio_country ~ elevation + as.factor(convex) + 
    x * y, data = .)

Residuals:
     Min       1Q   Median       3Q      Max 
-0.30581 -0.13120 -0.01994  0.11547  0.49772 

Coefficients:
                    Estimate Std. Error t value Pr(>|t|)   
(Intercept)         0.245177   0.079120   3.099  0.00209 **
elevation           0.008006   0.004707   1.701  0.08977 . 
as.factor(convex)1  0.009696   0.019665   0.493  0.62228   
x                  -0.069439   0.032596  -2.130  0.03380 * 
y                   0.004941   0.032256   0.153  0.87833   
x:y                 0.014815   0.014584   1.016  0.31035   
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 0.1705 on 372 degrees of freedom
Multiple R-squared:  0.08478,   Adjusted R-squared:  0.07248 
F-statistic: 6.892 on 5 and 372 DF,  p-value: 3.618e-06
touch_model_country %>% #filter(country!="Switzerland") %>%
lm(formula = touchratio_center ~ elevation + as.factor(convex) + x*y) %>% summary()

Call:
lm(formula = touchratio_center ~ elevation + as.factor(convex) + 
    x * y, data = .)

Residuals:
     Min       1Q   Median       3Q      Max 
-0.59725 -0.24964 -0.05573  0.23752  0.78698 

Coefficients:
                    Estimate Std. Error t value Pr(>|t|)    
(Intercept)         0.044263   0.141648   0.312 0.754848    
elevation           0.028905   0.008427   3.430 0.000671 ***
as.factor(convex)1 -0.063228   0.035206  -1.796 0.073316 .  
x                   0.032725   0.058356   0.561 0.575281    
y                   0.215671   0.057747   3.735 0.000217 ***
x:y                -0.069896   0.026110  -2.677 0.007758 ** 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 0.3053 on 372 degrees of freedom
Multiple R-squared:  0.1533,    Adjusted R-squared:  0.1419 
F-statistic: 13.47 on 5 and 372 DF,  p-value: 4.41e-12

4. To what extent are touch areas correlated with number of words spoken by the participant?

There does not seem to be any correlation between words spoken by the participant and touch ratios in either model.

touch_words_condition %>% ggplot() + aes(x = words_P, y = touchratio , color = as.factor(convex)) + geom_point() 

There is however a correlation between words spoken by the participant and by the experimenter:

touch_words_condition %>% ggplot() + aes(x = words_I, y = words_P , color = as.factor(convex)) + geom_point() 

5. Do Participants speak while touching, or either speak or touch?

Laura separate analysis

6. Is there a correlation between amount of touch and the reasoning by the Participant?

Correctness, confidence, uncertainty. Which areas and features do participants feel uncertain about? For example, identifying categories was really tricky with the model, and made people unsure about the data

7. Qualitative analysis of uncertainty and touch behavior

Discussion of parts of the model that are very frequently touched, and takeaways about the design of the physicalization.

8. Where are people touching?

Which different kinds of touches do we observe, and how do they impact reasoning? Analysis Bernice.