IDSC 4210 Midterm Review

One goal of data visualization

reduce the “time to insight”

For Data Analysts, Visualization helps you in..

• Exploring data structure

• Detecting Outliers and unusual Groups

• Identifying trends and clusters

• Presenting your results

Data Visualization allows you to

• Simplify complex information (even from a large datasets)

• Enhance Decision Making Process

Steps in Creating Data Visualization

1. What is your point: identify business objective (hypothesis)

2. How can you emphasize your point in your graph: identify data types, then choose graph type (proof)

3. What does the final graph show exactly: aesthetic mapping (color, shape, size) (explaining)

What is the 15 second rule?

Data visualizations should convey meaning in 15 seconds or less.

How quickly is a first impression made

7 seconds or less

select()

select variable

filter()

select rows

arrange()

order rows by variable

mutate()

create new variable

summarize()

aggregate the data

group_by()

group rows by variable

starts_with()

select based on prefix

ends_with()

select based on suffix

num_range()

select based on prefix & numeric range

contains()

matches a string within the variable

matches()

more general matching using regular expressions

one_of()

selects columns from a group of names

Filter(data, logical test)

==, >=, <=, !=(not equal too)

%>%

pipes: used to chain commands

desc()

sort variable in descending order

Pre-attentive attributes

generally the best way to present data because we can see these patterns without thinking or processing

What do pre-attentive attributes do?

• Draw Users Attention to the intended place

• create visual hierarchy

Pre-attentive attributes for bar charts

Form (length, width), color

Pre attentive attributes for line Graphs

Orientation, length, curvature, color, shape

Scatter Plots

2D position, color, form (shape, size)

What is the most important color consideration in data visualization?

Contrast

Color Best Practice

Limit number of colors

What is the 30, 60, 90 rule

 use 60% of a neutral color, 30% of a supplementary color, 10% of a color that pops 

Color in Mobile apps?

you can just use one single primary color with different brightness and saturations

Sequential Color in data visualization

data is ordered from high to low

Diverging Color

two sequential colors with a neutral midpoint

Categorical Color

contrasting colors for individual comparison

Highlight Color

color used to highlight something

Alert Color

Color used to get readers attention

Tufte’s 3 design principles

1. Graphical integrity: graph must be honest portrayal of data

2. Maximize Data to Ink Ratio

3. Chart Junk: all elements must be necessary

4 data types

• Nominal/Categorical Data

• Ordinal Data

• Interval Ratio

• Ratio Data

Nominal (categorical) Data

• Data placed in categories according to a specified characteristic 

• Categories bear no quantitative relationship to one another 

Ordinal (data)

• Data that is ranked or ordered according to some relationship with one another 

• No fixed units of measurement (the athlete that wins gold could be way better than the silver but in terms of rank it won’t matter) 

Interval Data

• Ordinal data but with constant differences between observations 

• No true zero point - 0 degrees does not mean there is an absence in temperature 

• Ratios are not meaningful

Ratio Data

• Continuous vales and have a natural, meaningful zero point

• Ratios are meaningful

In tableau what data types are dimensions

Nominal (categorical) and Ordinal

In tableau which data would be considered a measure?

Interval and Ratio Data

What are line graphs used to convey

TREND DATA

y is a measure, x is a time series (dimmension)

what are scatterplots used to convey

CORRELATIONS

x and y are measures

What are bar graphs used to convey?

MEASURES ACROSS DIMENSIONS

y is measure

x is a dimension

Histogram- distribution of measure

aes()

Aesthetic attributes

geoms()

Geometric Objects

stats()

statistical transofrmation

coord()

coordinates

facet()

facets

label()

with labels

Design Thinking

a philosophy and toolkit that helps you solve
problems through a creative, and human-centered lens.

What is Step 1 in design thinking?

Empathize: Understand people/organizations, within the given context.
It is your effort to understand the way they do things

What is step 2 in design thinking?

Define: Define the problem you are taking on, based on what you have
learned about your user and about the context in Step 1

What is step 3 in design thinking?

Ideate: Suggest ideas that can help organizations to find solution/insight
(Hypotheses Generation).

What is step 4 in design thinking?

Prototype: Generate artifacts that can provide answer/insight to solve the
problem

What is step 5 in design thinking?

Test: Refine the prototype.

How do you apply design thinking to data visualization?

alpha()

a solution to overplotting, adds transparency to the data points, helping visualize overlapping data

What does this code return?
prvfc + geom_point(alpha=.6, aes(color=QTY))

What does the following code return? prvfc + geom_point(alpha=.6,
aes(color=QTY))+scale_color_distiller(palette = "Spectral", name =
"#Tickets")

What does this code return? prvfc + geom_point(alpha=.3,
aes(size=QTY))+scale_size(breaks=c(1,5,15))

what does this code return? tpg+geom_bar(stat="summary",fun="mean",aes(fill=Game),position="dodge"
)+scale_fill_manual(values=c(Premium="Green",Regular="Orange",Value="R
ed"))+theme(axis.text.x = element_text(angle=45,hjust=1,size=6))

What does this Code return? prvfc + geom_point(alpha=.6,
aes(color=QTY))+scale_color_gradient(low="Pink",high="Purple")+
facet_grid(~Game)

What does this code return:pg+geom_bar(stat="summary",fun="mean",aes(fill=Game),position="dodge")+t
heme_bw()+coord_flip()

How many categories can you compare in a bar graph?

No more than 7 categories

How many categories can a horizontal bar chart have?

7-15 categories sorting ascendingly.

What is a correlation?

a statistical measure that indicates the
strength and direction of the relationship between two
quantitative variables, ranging from -1 to 1

What does a correlation of zero mean

no correlation, random assortment of points

what does a correlation of 1 or -1 mean?

perfect correlation all points on a single line

what is the magnitude of slope of a line?

slope coefficient y= mx+b

R-squared (coefficient of determination)

• % of the variability in dependent variables Y explained by the independent variable X 

• In essence, it shows how well X explains the variation in Y 

• If R^2 = 0.75, 75% of the variation in Y can be explained by X 

• The remaining 25% is due to other factors 

Tricks for over plotting

• First and foremost, check the stats:
correlation and regression
– This tells us about the relationship
• Take a sub sample
• Gradients (use alpha = )
• No fill
• Jittering
• Faceting

What will this code return?
aialpha<- exp %>% ggplot(aes(x=age, y=Income))+geom_point(alpha=.3)
aialpha

gradient

What does this code return? expsample <- sample_n(exp,1000)
aisam <- expsample %>% ggplot(aes(x=age, y=Income))+geom_point()+geom_smo
()
One trick to overcoming over-plotting is
sampling: quick and easy!!!

Sampling

What does this code return:aismall<- exp %>% ggplot(aes(x=age, y=Income)) +
geom_point(shape=21)

no fill (often the best choice)

What does this code return: aijitter<- exp %>% ggplot(aes(x=age,
y=Income))+geom_jitter(height=8,width=8,alpha=.3)

Jittering

What does this code return: gexpf<-mutate(exp,sex=ifelse(gender==0,"F","M"))
aig <- gexpf %>% ggplot(aes(x=age, y=Income))+
geom_point(alpha=.3)+facet_grid(.~sex)

Matrix

What part to whole pie chart is generally not ever the best choice?

Pie Chart

1. Hard perceive area

2. Hard to show multiple data points

3. Hard to perceive angles

When there are 2 or mare categories and we want to show parts-to-whole what graph should we consider?

Stacked bar chart

If you want to use a stacked column chart with time intervals what do you have to do.

Make sure dates have the same time interval

What does this code return: ggplot(tk, aes(x = reorder(Opponent, Opponent, function(x) -length(x)),
fill = Reseller)) + geom_bar(position = "stack") + labs(title = "Ticket Sales by
Opponent and Reseller", x = "Opponent", y = "Number of Transactions", fill =
"Reseller")

What does this return:orprice<-tk %>% ggplot(aes(x=reorder(Opponent,Price),y=Price)) + geom_bar(stat="summar
()

When looking at a time series what visualizations should we consider?

Line graph, bar graph, dot graph

When do you use dot plans?

when you have irregular intervals of time,
and/or when connection of dots is inappropriate