Are You Down With L.A.V?

But Did It Work?

Time Create Some New Variables

As fun as it is to explore the variables we already have in the data frame we need to move on to creating some of our own. One thing this data frame doesn't tell us is how many points each team scored during the course of the season. I also see a metric that captures the number of points per field goal attempt. Seems like a daunting task, it would be in Excel but we hate Excel and don't need it when we have nuclear weapons like R. Yet again there are countless ways to achieve both of these things in R but my favorite way is to use dplyr's mutate function.

all_nba_team_data %>>%
    filter(!is.na(g)) %>>%
    mutate(points_total = g * pts.g,
                 points_per_fga = points_total / fga) -> all_nba_team_data
all_nba_team_data %>>% 
    select(team,season,points_total, points_per_fga) #make sure it worked

## Source: local data frame [1,362 x 4]
## 
##                      team    season points_total points_per_fga
## 1        Dallas Mavericks 2014-2015        963.9       1.276689
## 2  Portland Trail Blazers 2014-2015        948.6       1.224000
## 3         Toronto Raptors 2014-2015        948.6       1.278437
## 4            Phoenix Suns 2014-2015        838.4       1.243917
## 5   Golden State Warriors 2014-2015        838.4       1.332909
## 6          Boston Celtics 2014-2015        732.2       1.190569
## 7           Brooklyn Nets 2014-2015        831.2       1.272894
## 8        Sacramento Kings 2014-2015        934.2       1.326989
## 9           Chicago Bulls 2014-2015        932.4       1.307714
## 10        Houston Rockets 2014-2015        826.4       1.341558
## ..                    ...       ...          ...            ...

Cool like Joe Johnson at end of a game with the Brooklyn Nets looking for the win, but we went to all that work to add those new columns we should at least explore them a little. How about trying to find which team had the highest ever points per field goal attempt?

all_nba_team_data %>>%
    filter(max(points_per_fga) == points_per_fga) %>>%
    select(team, season_end, points_per_fga, playoff_team)

## Source: local data frame [1 x 4]
## 
##        team season_end points_per_fga playoff_team
## 1 Utah Jazz       1995       1.376369         TRUE

Well, well, well those 1994-95 Utah Jazz. That was a pretty good team with superstars like Felton Spencer, John Crotty, Blue Edwards, Adam Keefe and this bad ass at age 35.

I don’t know, maybe that team had a few other decent players, just can’t think of them right now! This whole analysis thing has been fun and we just scratched the surface of what R can do but we came to here visualize, let’s do it.

Visualize What You Can’t C

No better way to think about it than those lyrics from the late Tupac Shakur. We can see the data of course, all 44,946 pieces of it, but its hard to really see what’s going on in a tabular format. That, ladies and gentleman, is what data visualization is for.

What Do We Visualize?

Before we get started visualizing we need to decide what exactly we want to explore. I keep hearing about this Daryl Morey character and how he is all into all these crazy things including data science. I also hear rumors that he is encouraging a game-plan centered around shooting tons of three pointers and even tried to formulate his roster around doing that.

Can data analysis tell us anything about this and the history of the three point shot? Of course it can.

Step 1: Filter Down to the Three Point Era

First we need to create a data frame that includes only the seasons since the advent of the Three Point Shot Era. We could use Google to find the answer but we are learning we are data scientists so let’s use R to answer the question of what years cover this era. How can R do this? When we looked at the summary of our data frame we noticed that X3p., x3p and X3pa columns all contained NA values. Hmmm, maybe thats the answer, if we filter out NAs in one of those columns intuition says that should work.

all_nba_team_data %>>%
    filter(!is.na(X3pa)) -> era3pt_shot #create the 3pt era data frame by filter  out NA remember that that symbol ! means NOT and is.na is a function to find NA values

It Worked, JEAH

Looks like R got us the right answer and if you don’t trust R, this should do the trick.

Step 2: Add A New Discrete Columns for the Decade

We have one item before to take care of before getting to the best part and that’s adding a discrete column for the decade which we intend utilize to add some spice to our visualizations..

era3pt_shot$season_end %>>% #select season end
    cut(breaks = c(0,1989,1999,2009,2015), #use cut to tell it  end of each decade
            labels = c('1980s','1990s','2000s','2010s') #label the new column
            ) -> era3pt_shot$decade

Step 3A: Get Hyped

Step 3B: Visualization Time

Alright we have what we need, its #DataViz time.

Let’s look at a colored Scatter Plot of Total Three Point Shot Attempts against Total Field Goal Attempts. Note since these variables are not calculated in real terms across the seasons [remember our 2015 season data is on going] we must filter the 2014-15 data in order to keep things consistent and not ruin the visualization.

Time to Admire Our First Visualization

era3pt_shot %>>%
    filter(!season_end == 2015) %>>% #filter out 2015
    ggvis(x =~fga, y =~X3pa, fill=~factor(decade), shape=~factor(decade), fillOpacity := 0.65) %>>% #fga vs 3pt attempts colored by decade
    layer_points() %>>% #ggvis nomenclature for scatter
    add_axis("x", title = "Field Goal Attempts", 
                     title_offset = 50) %>>% #fix x title
    add_axis("y", title = "3PT Shot Attempts",
                     title_offset = 55) %>>% #fix y title
    add_legend(c("fill",'shape'), 
                         title = "Decade") #clean up the legend

Look at that beautiful ggvis scatter plot. What do we see here? Well first it looks like back in the 1980s and 1990s team’s on average used to take significantly more field goal attempts, we even see one insane out-lier, the absolutely DREADFUL 1990-91 Denver Nuggets who attempted an insane 8,668 field goals [bonus challenge to those up for it, try to use the 2 lines of dplyr I used to figure this out].

In addition to that it, looks like as we move through the decades, despite less overall field goal attempts, teams appear to be taking more 3 point shots. There are all sorts of factors that may have influenced this pattern outside of the magical powers of Daryl Morey, I don’t want to give it away but if you have time, try to investigate how the NBA has changed the Three Point Shot rules since its adoption in the 1979-80 season. Though we won’t get into it in this post, it looks like the data during the 1990s and the 2010s would be ripe for Clustering Analysis [hint there maybe actual clusters in what we see or these very apparent clusters data maybe the result of different external force].

This visualization is quite interesting and the more we look and think about it the more potential investigative topics may come across but this visualization isn’t fair, it's missing the 2015 season data and it isn't easy to see if there have been trends in how the 3 point shot progressed as the years have passed.

era3pt_shot %>>%
    mutate(X3pa_per_min = X3pa/mp) -> era3pt_shot #add the new variable

Step 5: Time to Visualize It, the Iggy Azelea Way

Ok we have the variables we need let’s get to it. For this visualization we want to look at 3 Point Shot Attempts Per Minute by Season End.

We could do this visually in any number of ways but since I want to get fancy and showcase the power of easy regression analysis in ggvis thanks to the hard work of the man, the myth, the legend Hadley Wickham we are going to stick with a scatter plot and layer on a regression line.

era3pt_shot %>>%
    ggvis(x =~season_end, y =~X3pa_per_min, fill = ~factor(decade), fillOpacity := 0.45) %>>%
    add_axis("x", title = "", values = seq(from = 1980,to =2015,1), format="####", title_offset = 50,
                     properties = axis_props(
                        ticks = list(stroke = "black"),
                        majorTicks = list(strokeWidth = 2),
                        labels = list(
                            angle = 50,
                            fontSize = 11,
                            align = "left",
                            baseline = "middle",
                            dx = 3
                        ))) %>>%
    add_axis("y", title = "3PT Point Attempts Per Minute",
                     title_offset = 55, ticks = 20) %>>%
    layer_points(shape=~factor(decade)) %>>%
    group_by(decade) %>>%
    layer_model_predictions(model = 'lm', stroke =~factor(decade), se  = T) %>>%
    add_legend(c("fill","stroke","shape"),orient = 'right', title = "Decade",
                         properties = legend_props(
                            title = list(fontSize = 12),
                            labels = list(fontSize = 10, dx = 10),
                            symbol = list(stroke = "black", strokeWidth = 1,
                                                        size = 100)
                         ))

Damn That 1362 Point Plot is FINE

Look at this amazing regression scatter plot highlighting every single 3 point shot attempt per minute since the introduction of the shot. It appears there was certainly a an upward trend from the 1980’s into the late 1990s and another starting in the mid 2000s through this season. There are a host of potential explanatory factors and other follow on analysis we could perform based on this viz, some which may be topics in forthcoming posts, but before we call it a day we want to look at, and visualize, one last item.

When looking at this graph it may be should be come apparent that, with only a few exceptions, each season’s leader in three point shot attempts appears to be higher than the prior season’s leader, signifying a potential trend. In fancy statistical words, this possible trend appears to display linearity. Appearances can deceive though and we need to thoroughly investigate this hypothesis in keeping with the data ninja code of mathematically proving the possibility of a relationship.

There are a number of ways to do this while adhering to this code, all of which R or any good programming language makes easy for you. In this case we are going to use the results of a linear model, the easiest of which to do in R uses the lm function.

Step 6: Let’s Do It, Linear Regression That Is.

era3pt_shot %>>% 
    group_by(season_end) %>>% #group by season
    filter(X3pa_per_min == max(X3pa_per_min)) %>>%  #take the max of each year
    (lm(X3pa_per_min ~ season_end, data = .)) %>>% #apply lm against season
    summary #let's look at the summary data to see the fit

## 
## Call:
## lm(formula = X3pa_per_min ~ season_end, data = .)
## 
## Residuals:
##       Min        1Q    Median        3Q       Max 
## -0.019115 -0.006928 -0.000761  0.002920  0.032767 
## 
## Coefficients:
##               Estimate Std. Error t value Pr(>|t|)    
## (Intercept) -6.5236007  0.3623104  -18.01   <2e-16 ***
## season_end   0.0033032  0.0001814   18.21   <2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.01131 on 34 degrees of freedom
## Multiple R-squared:  0.907,  Adjusted R-squared:  0.9043 
## F-statistic: 331.7 on 1 and 34 DF,  p-value: < 2.2e-16

Good Job Intuition, We Were Right That’s a 90% Fit to the Data

Our instincts served us well. There is unquestionably a linear relationship over time that indicates that the team that leads the league in 3 point shot attempts per minute has been going up year by year. What this means for the future we can’t say definitively and I by no means would extrapolate that this trend will continue, but historically it has. Now for our final visualization let’s look at who the teams were that actually lead the league in three point shot attempts per minute were.

Step 7: Create A Better “Real” Variable to Plot With

Maybe, but I think we could do better. We know that there are 5 players on the court per team at a time, that there are a minimum of 4 quarters each of 12 minute length. That means, excluding overtime, there are 240 available minutes per game. So we can take our new variable, and multiply it by 240 to get a new version of it that contains the same information but per 240 minutes, essentially it is a roundabout that should nearly mimic three shot attempts per game if we were to calculate it. Let’s add our new variable and then create a new data frame containing only the top performing teams by season.

era3pt_shot %>>%
    mutate(X3pa_per_240_min = X3pa_per_min * 240) %>>%
    group_by(season_end) %>>%
    filter(X3pa_per_min == max(X3pa_per_min)) %>>%
    select(season_end, team, X3pa_per_240_min) -> top_teams

Step 8: Lets Visualize These Teams

Now that we have data and new variable this let’s quickly visualize it, but since I am stickler for colors matching the entities they are associated with I want to find the right colors for each team. Fortunately I’ve already done this and we can just read in the file with this data and add in the 2 teams missing teams [the San Diego Clippers and the late Seattle Supersonics, RIP].

#Bring in Correct Colors
'https://asbcllc.com/data/NBA/team_colors.csv' %>>% read.csv %>>% data.frame %>>%
    tbl_df -> active_colors
data.frame(team = top_teams$team %>>% unique) %>>% tbl_df %>>%
    arrange((team)) -> teams
teams %>>% merge(active_colors,all.x = T) -> teams
'#EE2944' -> teams[14,2] #add the San Diego Clippers
'#266A2E'-> teams[15,2] #add the Sonics

We are now ready to create our final visualization.

top_teams %>>%
    ggvis(x =~season_end, #season
                y =~X3pa_per_240_min,#by our 240 stat showi
                text := ~team, #plot the team name
                fill = ~team) %>>% #color the team
    add_axis("x", title = "",  #no title
                     values = seq(from = 1980,to =2015,1), #plot the years
                     format="####", title_offset = 50, #format the axis
                     properties = axis_props(
                        ticks = list(stroke = "black"),
                        majorTicks = list(strokeWidth = 2),
                        labels = list(
                            angle = 50,
                            fontSize = 11,
                            align = "left",
                            baseline = "middle",
                            dx = 3
                        ))) %>>%
    add_axis("y", title = "3PT Attempts Per 240 Available Minutes", 
                     format="####", #uses d3 axis format
                     title_offset = 55, ticks = 20) %>>% #fix title and ticks
    layer_text() %>>% #adds the text 
    scale_nominal(property = 'fill', #use our pretty colors
                                domain = as.character(teams$team),
                                range = as.character(teams$primary_color)) %>>%
    add_legend("fill", orient = 'right', 
                         title = "Team", #need a pretty legend
                         properties = legend_props(
                            title = list(fontSize = 12),
                            labels = list(fontSize = 10, dx = 10),
                            symbol = list(stroke = "black", strokeWidth = 1,
                                                        size = 100)
                            ))

Wow look at this amazing chart. This clearly validates the linearity we explored earlier and shows that in recent times this Daryl Morey guy has clearly pushed for and structured teams around taking lots of three point shots. What is even more interesting is that there appears to be another candy loving, last minute ETO signing, father of 8 that is nearly universally despised by Nets and Lakers fans whose presence on a roster, whether by design or chance, appears as a constant in all but 1 of the team’s that have lead the NBA in 3 point shot attempts per 240 minutes since the 2009-10 season. Any idea who he may be? Here’s a hint