Investigate a Dataset

Baseball All-Star Data

Justin Ibanez

Introduction¶

For my project I decided to investigate the baseball data. The question I asked was, "How are the all-stars different from the entire player base?" I looked at two characteristics to investigate this: birth country (or state for American born players) and handedness for batting and throwing.

Data Wrangling¶

The data for this project was taken from SeanLahman.com. The CSVs investigated were AllstarFull.csv, Master.csv and Appearances.csv. I loaded the CSVs into dataframes using Pandas and printed their heads to get an idea of their structure.

import pandas as pd

# read CSVs and print their heads
allstar_df = pd.read_csv('AllstarFull.csv')
master_df = pd.read_csv('Master.csv')
appearances_df = pd.read_csv('Appearances.csv')

allstar_df.head()

master_df.head()

appearances_df.head()

I started by looking for missing values starting with the all-stars dataframe.

# list number of missing values in allstar_df
allstar_df.isnull().sum()

playerID          0
yearID            0
gameNum           0
gameID           49
teamID            0
lgID              0
GP               19
startingPos    3489
dtype: int64

The gameID, GP and startingPos columns had missing values. The documentation included with this data said that startingPos would only have a value if the player was a starter, so it makes sense for there to be so many missing values. There was nothing explaining the other two columns, so I investigated them further.

# find entries with missing gameID
allstar_df.loc[pd.isnull(allstar_df['gameID'])]

All these entries were for 1945 all-star game. This game was cancelled due to World War II, which explains the missing values. I decided to keep these entries as these players were still chosen to be all-stars even if they did not play the game.

# find entries with missing GP
allstar_df.loc[pd.isnull(allstar_df['GP'])]

These entries corresponded to the 2015 all-star game. After doing a little research it seemed like these entries corresponded to players who were selected as all-stars, but did not appear in the game for some reason (such as injuries). I was unable to find a reason for every player (nor was every player who was replaced due to injury on the above list), so I was not completely confident in this explanation.

It was also noteworthy that these entries were the last 18 entries in the dataframe although I could not find a reason for the missing values from that angle.

I decided to keep these entries as all the players are listed on the roster,.

I moved on to the master dataframe.

# list number of missing values in master_df
master_df.isnull().sum()

playerID           0
birthYear        143
birthMonth       315
birthDay         464
birthCountry      73
birthState       626
birthCity        199
deathYear       9510
deathMonth      9511
deathDay        9512
deathCountry    9517
deathState      9569
deathCity       9521
nameFirst         39
nameLast           0
nameGiven         39
weight           871
height           805
bats            1191
throws           978
debut            193
finalGame        193
retroID           54
bbrefID            1
dtype: int64

Only playerID did not have missing values. Since this data goes back to 1820 the missing data was likely due to the information simply being unavailable. Furthermore, there would be no death time or place information for players who were still alive, which explains why there are significantly more missing values in those columns than the others. There was not much I could do to remedy the missing data, so I decided to continue using it as is.

The variables I were investigating had a relatively small amount of missing data points, but they would be affected by the missing data nonetheless. The most significant was the batting handedness where about 6% of the data was missing. It could also be possible for the missing birth country data to significantly affect the results if those players came from countries with low player contributions.

Finally, I looked for missing data in the appearances dataframe.

# list number of missing values in appearances_df
appearances_df.isnull().sum()

yearID           0
teamID           0
lgID           737
playerID         0
G_all          203
GS           51921
G_batting        0
G_defense      203
G_p              0
G_c              0
G_1b             0
G_2b             0
G_3b             0
G_ss             0
G_lf             0
G_cf             0
G_rf             0
G_of             0
G_dh         51718
G_ph         51718
G_pr         58006
dtype: int64

The columns lgID (league), G_all (total games played), GS (games started), G_defense (games in which the player appeared on defense), G_dh (games as designated pitcher), G_ph (games as pinch hitter) and G_pr (games as pinch runner) had missing data. Since G_defense, G_dh, G_ph and G_pr referred to positions, it made sense for there to be missing values as not every player would play in those positions in every season. Furthermore, GS would have missing data if a player did not start that season. However, lgID and G_all did not have a clear reason for missing data and needed to be investigated further. I started with with the league information.

# find entries with missing lgID
appearances_df.loc[pd.isnull(appearances_df['lgID'])].head()

appearances_df.loc[pd.isnull(appearances_df['lgID'])].tail()

These entries were from the years 1871 to 1875. Since the American League (the older of the two leagues that comprises the MLB) was founded in 1876, any entries prior to 1876 would not have a league ID. I looked at the games played information next.

# find entries with missing G_all
appearances_df.loc[pd.isnull(appearances_df['G_all'])].head()

All of these entries were from the 2013 season, but I could not find a reason why this data was missing. I was using the appearances data to help collect information on how handedness changed over time, so none of this missing data affected by analysis. I decided to ignore it.

I then cleaned up the inconsistencies in the country names in the master dataframe.

# make all country names consistent
def country_corrector(country):
    if country == 'CAN':
        return 'Canada'
    elif country == 'D.R.':
        return 'Dominican Republic'
    elif country == 'P.R.':
        return 'Puerto Rico'
    elif country == 'V.I.':
        return 'US Virgin Islands'
    return country

master_df['birthCountry'] = master_df['birthCountry'].apply(country_corrector)

The all-stars dataframe contained an entry for every appearance a player made in an all-star game. This means the same player will have multiple entries if he played in more than one all-star game in his career. I used the all-stars dataframe to create an array of all-star players and then used this list to extract the all-stars from the master dataframe.

# all-stars array
unique_allstars = allstar_df['playerID'].unique()

# all-star players
allstars = master_df.loc[master_df['playerID'].isin(unique_allstars)]

I noticed a discrepancy in this data. The unique_allstars array had 1741 elements. However, the allstars dataframe had 1740 rows. I found the data point in unique_allstars, but not in allstars.

# discrepancy
print len(unique_allstars) # 1741
print allstars.shape[0] # 1740

# find discrepancy
for player in unique_allstars:
    if player not in allstars['playerID'].tolist():
        print player # carperma01
        
# print discrepancy
allstar_df.loc[allstar_df['playerID'] == 'carperma01']

1741
1740
carperma01

It was not immediately clear to me what was wrong with this data point. After a bit of research I found the answer. All the player IDs were 9 characters long and 'carperma01' contained 10 characters. This was due to a typo as 'carperma01' should have been 'carpema01'. This data point refered to Matt Carpenter's second all-star selection. I corrected this discrepancy and recreated unique_allstars and allstars.

# correct discrepancy
allstar_df.set_value(4965, 'playerID', 'carpema01')

#recreate array and dataframe
unique_allstars = allstar_df['playerID'].unique()
allstars = master_df.loc[master_df['playerID'].isin(unique_allstars)]

# double check that values match
print len(unique_allstars) # 1740
print allstars.shape[0] # 1740

1740
1740

With the discrepancy cleared I organized the player data. Since America had a much larger number of players than any other country they would have dominated most measurements. Therefore, I decided to look at them separately. I wanted to create a dataframe containing all-star and player populations for American and non-American players.

I separated the all-stars and players (from the master dataframe) into American and non-American born players. I grouped each of the resultant dataframes by birth state/country and counted to get the all-star and player populations for each state/country. After renaming some columns for clarity, I concatenated these columns to get my desired dataframes. I also computed the percentage of all-stars in each dataframe for future analysis.

# divide all-stars and players into American and non-American
american_allstars = allstars.loc[allstars['birthCountry'] == 'USA']
non_american_allstars = allstars.loc[allstars['birthCountry'] != 'USA']
american_players = master_df.loc[master_df['birthCountry'] == 'USA']
non_american_players = master_df.loc[master_df['birthCountry'] != 'USA']

# group by state/country
american_allstars_by_state = american_allstars.groupby('birthState').count()
non_american_allstars_by_country = non_american_allstars.groupby('birthCountry').count()
american_players_by_state = american_players.groupby('birthState').count()
non_american_players_by_country = non_american_players.groupby('birthCountry').count()

# clean up column names to prepare for concatenation
american_allstars_by_state['allStars'] = american_allstars_by_state['playerID']
non_american_allstars_by_country['allStars'] = non_american_allstars_by_country['playerID']
american_players_by_state['players'] = american_players_by_state['playerID']
non_american_players_by_country['players'] = non_american_players_by_country['playerID']

# concatenate
american_players_by_state = pd.concat(
    [american_players_by_state['players'], american_allstars_by_state['allStars']], axis=1)
non_american_players_by_country = pd.concat(
    [non_american_players_by_country['players'], non_american_allstars_by_country['allStars']], axis=1)

# replace NaN with 0
american_players_by_state.fillna(0, inplace=True)
non_american_players_by_country.fillna(0, inplace=True)

# add percentAllStars column
american_players_by_state['percentAllStar'] = (american_players_by_state['allStars']
                                               / american_players_by_state['players']) * 100.00
non_american_players_by_country['percentAllStar'] = (non_american_players_by_country['allStars']
                                                     / non_american_players_by_country['players']) * 100.00

# print final dataframes
american_players_by_state.head()

non_american_players_by_country.head()

The next step was to organize the data for handedness. I started by adding the handedness data from the master dataframe to the all-stars and appearances dataframes. I limited the player database to the same years as the all-stars.

# handedness of all players
handedness = master_df[['playerID', 'bats', 'throws']]

# add handedness to all-stars
allstar_handedness = allstar_df.merge(handedness, on='playerID', how='left')[['playerID', 'yearID', 'bats', 'throws']]

# print result
allstar_handedness.head()

# add handedness to player appearances
player_handedness = appearances_df.merge(handedness, on='playerID', how='left')[['playerID', 'yearID', 'bats', 'throws']]
player_handedness = player_handedness.loc[player_handedness['yearID'] >= 1933]

# print result
player_handedness.head()

I grouped the all-stars and players by year and used the value_counts method (as a proportion) to create four series containing different types of handedness: all-star batting, all-star throwing, player batting and player throwing.

# all-star batting
allstar_bat_handedness_by_year = allstar_handedness.groupby('yearID')['bats'].value_counts(normalize=True)
allstar_bat_handedness_by_year = allstar_bat_handedness_by_year.fillna(0)

# all-star throwing
allstar_throw_handedness_by_year = allstar_handedness.groupby('yearID')['throws'].value_counts(normalize=True)
allstar_throw_handedness_by_year = allstar_throw_handedness_by_year.fillna(0)

# player batting
player_bat_handedness_by_year = player_handedness.groupby('yearID')['bats'].value_counts(normalize=True)
player_bat_handedness_by_year = player_bat_handedness_by_year.fillna(0)

# player throwing
player_throw_handedness_by_year = player_handedness.groupby('yearID')['throws'].value_counts(normalize=True)
player_throw_handedness_by_year = player_throw_handedness_by_year.fillna(0)

Exploration¶

I started by looking at the birth states/countries. I began by plotting the raw data. I decided to use bar graphs, because I was comparing the different amounts of all-stars and the difference in height would easily provide a visual indicator for these differences.

%pylab inline

import matplotlib.pyplot as plt
import seaborn as sns

Populating the interactive namespace from numpy and matplotlib

master_df['birthCountry'].value_counts().plot(kind='bar', figsize=[15,14])
plt.title('Raw Birth Country Statistics')
plt.xlabel('Country')
plt.ylabel('Number of Players')

<matplotlib.text.Text at 0xcae5f28>

USA completely dominated every other country, which justified my choice in looking at it separately. I plotted the raw birth state data next.

master_df['birthState'].value_counts().plot(kind='barh', figsize=[15,45])
plt.title('Raw Birth State Statistics')
plt.xlabel('Number of Players')
plt.ylabel('State')

<matplotlib.text.Text at 0xe871898>

Again, American states dominated the data. Furthermore, the sheer amount of states in the master dataframe also justified my choice to group them by the more general category of countries. In particular, most countries contributed players from only a few states within them, so it made sense not to look at them separately. Although the argument could be made to give the other countries with relatively large amounts of player contribution (such as Canada and the Dominican Republic) the same treatment as USA.

I moved on to the American all-stars and their state of birth.

american_players_by_state.describe()

From the description of the data there was an average of about 323 players per state. About 29 or 9.3% of those players were all-stars. There was a large difference between the minimum and maximum values and the 75th percentile was very small compared to the maximum.

american_players_by_state['allStars'].plot(kind='barh', figsize=[15,14])
plt.title('American All-Stars by Birth State', fontsize=18)
plt.xlabel('Number of All-Stars', fontsize=14)
plt.ylabel('Birth State', fontsize=14)

<matplotlib.text.Text at 0xeeef208>

What stuck out (literally) was that California had many more all-stars than the other states (about three times more than the state with the second most all-stars), which explained the statistics noted above. This could be attributed to their large population, so I also plotted the percentage of all-stars as well to reduce the effect of a large population.

american_players_by_state['percentAllStar'].plot(kind='barh', figsize=[15,14])
plt.title('American All-Stars by Birth State - Percentage', fontsize=16)
plt.xlabel('Percent of Player Base that are All-Stars', fontsize=14)
plt.ylabel('Birth State', fontsize=14)

<matplotlib.text.Text at 0x10c9fdd8>

This plot looked signficantly different from the previous one. California no longer had the largest value and in general the states' values were a lot closer together. The top three states were Idaho, Wyoming and Alaska. Maine had a low value in both plots.

While looking at the percentage did prevent large populations from dominating the plot, it also had the effect of overvaluing all-stars in states with smaller populations. A state with a small amount of players (such as Idaho, Wyoming and Alaska) only needed a small amount of all-stars to achieve a large percentage. However, there was no way of knowing if these all-stars were outliers and not indicative of the general player base. Therefore, in order to mitigate this effect I removed all states with less than 25 all-stars and plotted the percentages again.

to_plot = american_players_by_state.loc[american_allstars_by_state['allStars'] >= 25]
to_plot['percentAllStar'].plot(kind='barh', figsize=[15,14])
plt.title('American All-Stars by Birth State - Percentage', fontsize=16)
plt.xlabel('Percent of Player Base that are All-Stars', fontsize=14)
plt.ylabel('Birth State', fontsize=14)

<matplotlib.text.Text at 0x11f01eb8>

Only 21 states remained after imposing this condition. And the percentages of these remaing states were relatively close together compared to the previous plots. The state with the largest percentage here was Arkansas even though it had a relatively low amount of all-stars. This implied that Arkansas born players are disproportionately more likely to become all-stars. New York was the state with the third most all-stars, but it had a relatively low percentage. This implied that the New York born players are disproportionately less likely to become all-stars. The same could also be said for Pennsylvania. However, the relative closeness of the percentages means that birth states did not drastically affect a player's chance at being an all-star. That is to say, California contributes more all-stars due to their larger population and not due to Californians being inherently good at baseball.

Finally, I plotted the number of players per birth state to look at this from another angle.

american_players_by_state['players'].plot(kind='barh', figsize=[15,14])
plt.title('American Players by Birth State', fontsize=18)
plt.xlabel('Number of Players', fontsize=14)
plt.ylabel('Birth State', fontsize=14)

<matplotlib.text.Text at 0x121a9e48>

Once again, California dominates the plot. Overall, the shape of this plot mimicked the shape of the all-stars plot. States with large amounts of all-stars also had large amounts of players and vice versa. The differences between the proportions were more easily seen in the percentages plot, but in general the all-star plot was a microcosm of the players plot. This reinforced my inital conclusion that birth state does not dratstically affect all-star likelihood.

I moved on to the rest of the countries next starting with a description of the data.

non_american_players_by_country.describe()

The other countries contributed an average of about 44 players with an average of 5 all-stars. However, the average percentage of all-stars contributed was 7.5% (although 5/44 is about 11%). This implied a heavily skewed distribution. This is further supported by the 75th percentile for the players and all-stars being extremely low compared to the maximum value.

It would turn out that some countries contributed one player who would also go on to become an all-star leading to a 100% all-star rate.

I plotted the all-stars for each country (for countries with at least one all-star).

to_plot = non_american_players_by_country.loc[non_american_players_by_country['allStars'] > 0]
to_plot['allStars'].plot(kind='barh', figsize=[15,14])
plt.title('Non-American All-Stars by Birth Country', fontsize=18)
plt.xlabel('Number of All-Stars', fontsize=14)
plt.ylabel('Birth Country', fontsize=14)

<matplotlib.text.Text at 0x126f5048>

Several countried clearly dominated this plot. Again, I decided to look at percentages. This time I limited the results to countries with at least 10 all-stars (many countries had percentages over 50 without this limitation).

to_plot = non_american_players_by_country.loc[non_american_players_by_country['allStars'] > 10]
to_plot['percentAllStar'].plot(kind='barh', figsize=[15,14])
plt.title('Non-American All-Stars by Birth Country - Percentage', fontsize=18)
plt.xlabel('Percentage of Player Base that are All-Stars', fontsize=14)
plt.ylabel('Birth Country', fontsize=14)

<matplotlib.text.Text at 0x12857eb8>

Many of the countries were eliminated using this criteria and the countries that remained had percentages close to each other despite contributing largey disparate amounts of all-stars. This could be attributed to the data coming from American baseball leagues and so it made sense that foreigners would only compete if they had exceptional skill. A number of countries that contributed a very small number of players (such as two to five) would often have most of those players become all-stars.

Finally, I plotted the players per country.

non_american_players_by_country['players'].plot(kind='barh', figsize=[15,14])
plt.title('Non-American Players by Birth Country', fontsize=18)
plt.xlabel('Number of Players', fontsize=14)
plt.ylabel('Birth Country', fontsize=14)

<matplotlib.text.Text at 0x12918278>

This plot was consistent with the all-stars plot moreso than the American data. The countries with high amounts of all-stars also had high amounts of players and the countries that contributed few or no all-stars also contributed few players.

I looked at the handedness data next. I started by plotting the raw handedness data from the master database.

master_df['bats'].value_counts().plot(kind='bar')
plt.title('Raw Batting Statistics')
plt.xlabel('Handedness')
plt.ylabel('Number of Players')

<matplotlib.text.Text at 0x16449470>

master_df['throws'].value_counts().plot(kind='bar')
plt.title('Raw Throwing Statistics')
plt.xlabel('Handedness')
plt.ylabel('Number of Players')

<matplotlib.text.Text at 0x16507080>

The main distinction between the plots was that players were ambidextrous in batting, but not in throwing. This suggested that players were more versatile at batting. There were more right-handed than left-handed players as well, which was expected.

I moved on to the wrangled data to see how the handedness changed over time.

allstar_bat_handedness_by_year.unstack(level=1).describe()

C:\Users\justin\Anaconda2\lib\site-packages\numpy\lib\function_base.py:3834: RuntimeWarning: Invalid value encountered in percentile
  RuntimeWarning)

The all-star batting handedness could be broken down to about 63% right-handed, 30% left-handed and 7% ambidextrous. The values had a relatively large range. There was a little as 52% right-handed all-stars at one point, which was surprising given the raw data above.

I plotted the batting data for all-stars and players. Since I was dealing with a change over time I decided to use a line graph.

allstar_bat_handedness_by_year.unstack(level=1).plot(figsize=[10,8])
plt.title('All-Star Batting Handedness')
plt.xlabel('Year')
plt.ylabel('Percentage of Players')

<matplotlib.text.Text at 0x16852f28>

player_bat_handedness_by_year.unstack(level=1).describe()

The averages for the player base were similar to the all-stars, but it seemed the data did not vary as much.

player_bat_handedness_by_year.unstack(level=1).plot(figsize=[10,8])
plt.title('Player Batting Handedness')
plt.xlabel('Year')
plt.ylabel('Percentage of Players')

<matplotlib.text.Text at 0x16772d30>

What stood out to me was that the all-star plot was much more sporadic than the general player base, which supported the conjectures made above. I suspected that the general player base handedness would be steady due to the population not changing much from year to year. However, the sporadic nature of the all-stars implied that this was not the case for them. This could mean it was hard for someone to beomce an all-star multiple years in a row. There was also a slight increase in the ambidextrous players as time went on, which could mean the player base was becoming more versatile over time.

allstar_throw_handedness_by_year.unstack(level=1).describe()

About 20% of players were left-handed and 80% were right-handed when it came to throwing. These numbers varied by about 20% over the years, which further indicated the all-star handedness was more sporadic than the general player base.

allstar_throw_handedness_by_year.unstack(level=1).plot(figsize=[10,8])
plt.title('All-Star Throwing Handedness')
plt.xlabel('Year')
plt.ylabel('Percentage of Players')

<matplotlib.text.Text at 0x16a1ad68>

player_throw_handedness_by_year.unstack(level=1).describe()

The averages for the player base were similar to the all-stars. However, the varied by half as much as the all-stars, which further supported the trend of more stable statistics for the players compared to the all-stars.

player_throw_handedness_by_year.unstack(level=1).plot(figsize=[10,8])
plt.title('Player Throwing Handedness')
plt.xlabel('Year')
plt.ylabel('Percentage of Players')

<matplotlib.text.Text at 0x1810aa58>

Again, the sporadic nature of the all-stars relative to the player base further supported that it was hard for all-stars to have repeat showings. Furthermore, the difference in proportions was much greater for throwing than for batting. This was more proof that the players were better at batting than throwing. Overall, the plots reflected the conjectures made on the data's descriptions.

Conclusions¶

It is important to note that any relations implied by my analysis are not necessarily causal. There are many factors that could affect the results presented below. This project was focused on comparing all-stars to the general player base. However, all-stars are chosen by people whether it be the players, managers or fans rather than a competition or a player's stats. This means any trends in the all-stars' performance could be coincedental. For example, a popular player with average skill might have a better chance of becoming an all-star than a skilled, but unpopular player. Furthermore, being chosen to be an all-star once could affect your future performance. For example, maybe a player picked to be an all-star is more motivated and as a result performs better. Nevertheless, these are the conclusions I arrived to.

American all-star distribution by birth state was a microcosm of player distribution by birth state. States with a large number of players had a proportionately large number of all-stars and thus birth state does not drastically affect all-star likelihood.
Non-American countries sent their best players to compete in America, so they had a disproportionate amount of all-stars. By that I mean countries that contributed a small number of players would have most of those players become all-stars.
Players are more versatile at throwing than batting and their throwing versatility is increasing.
It was hard to become an all-star multiple years in a row

Acknowledgements¶

I would like to thank Caroline Buckey for creating the course on the data analyst process as well as numpy and Pandas, Mat Leonard for creating the course on Anaconda and Jupyter Notebook and Sean Lahman for providing the data.

	playerID	yearID	gameID	teamID	lgID	GP	startingPos
0	gomezle01	1933	ALS193307060	NYA	AL	1.0	1.0
1	ferreri01	1933	ALS193307060	BOS	AL	1.0	2.0
2	gehrilo01	1933	ALS193307060	NYA	AL	1.0	3.0
3	gehrich01	1933	ALS193307060	DET	AL	1.0	4.0
4	dykesji01	1933	ALS193307060	CHA	AL	1.0	5.0

	playerID	birthYear	birthMonth	birthDay	birthCountry	birthState	birthCity	deathYear	deathMonth	deathDay	...	nameLast	nameGiven	weight	height	bats	throws	debut	finalGame	retroID	bbrefID
0	aardsda01	1981.0	12.0	27.0	USA	CO	Denver	NaN	NaN	NaN	...	Aardsma	David Allan	220.0	75.0	R	R	2004-04-06	2015-08-23	aardd001	aardsda01
1	aaronha01	1934.0	2.0	5.0	USA	AL	Mobile	NaN	NaN	NaN	...	Aaron	Henry Louis	180.0	72.0	R	R	1954-04-13	1976-10-03	aaroh101	aaronha01
2	aaronto01	1939.0	8.0	5.0	USA	AL	Mobile	1984.0	8.0	16.0	...	Aaron	Tommie Lee	190.0	75.0	R	R	1962-04-10	1971-09-26	aarot101	aaronto01
3	aasedo01	1954.0	9.0	8.0	USA	CA	Orange	NaN	NaN	NaN	...	Aase	Donald William	190.0	75.0	R	R	1977-07-26	1990-10-03	aased001	aasedo01
4	abadan01	1972.0	8.0	25.0	USA	FL	Palm Beach	NaN	NaN	NaN	...	Abad	Fausto Andres	184.0	73.0	L	L	2001-09-10	2006-04-13	abada001	abadan01

	playerID	yearID	gameID	teamID	lgID	GP	startingPos
5050	santihe01	2015	NLS201507141	LAA	AL	NaN	NaN
5051	grayso01	2015	NLS201507142	OAK	AL	NaN	NaN
5052	bautijo02	2015	NLS201507143	TOR	AL	NaN	NaN
5053	millesh01	2015	NLS201507144	ATL	NL	NaN	NaN
5054	martica04	2015	NLS201507145	SLN	NL	NaN	NaN
5055	wachami01	2015	NLS201507146	SLN	NL	NaN	NaN
5056	rosentr01	2015	NLS201507147	SLN	NL	NaN	NaN
5057	hollima01	2015	NLS201507148	SLN	NL	NaN	NaN
5058	gordode01	2015	NLS201507149	MIA	NL	NaN	NaN
5059	stantmi03	2015	NLS201507150	MIA	NL	NaN	NaN
5060	scherma01	2015	NLS201507151	WAS	NL	NaN	NaN
5061	odayda01	2015	NLS201507152	BAL	AL	NaN	NaN
5062	papeljo01	2015	NLS201507153	PHI	NL	NaN	NaN
5063	burneaj01	2015	NLS201507154	PIT	NL	NaN	NaN
5064	boxbebr01	2015	NLS201507155	TBA	AL	NaN	NaN
5065	gordoal01	2015	NLS201507156	KCA	AL	NaN	NaN
5066	herreke01	2015	NLS201507157	KCA	AL	NaN	NaN
5067	cabremi01	2015	NLS201507158	DET	AL	NaN	NaN
5068	salech01	2015	NLS201507159	CHA	AL	NaN	NaN

	players	allStars	percentAllStar
birthState
AK	11	2	18.181818
AL	322	43	13.354037
AR	152	25	16.447368
AZ	99	7	7.070707
CA	2160	295	13.657407

	players	allStars	percentAllStar
Afghanistan	1	0.0	0.000000
American Samoa	1	0.0	0.000000
Aruba	5	0.0	0.000000
At Sea	1	0.0	0.000000
Australia	28	2.0	7.142857

	yearID	teamID	lgID	playerID	G_all	GS	G_batting	G_defense	G_c	...	G_2b	G_ss	G_lf	G_rf	G_of	G_dh	G_ph	G_pr
0	1871	BS1	NaN	barnero01	31.0	NaN	31	31.0	0	...	16	15	0	0	0	NaN	NaN	NaN
1	1871	BS1	NaN	barrofr01	18.0	NaN	18	18.0	0	...	1	0	13	4	17	NaN	NaN	NaN
2	1871	BS1	NaN	birdsda01	29.0	NaN	29	29.0	7	...	0	0	0	27	27	NaN	NaN	NaN
3	1871	BS1	NaN	conefr01	19.0	NaN	19	19.0	0	...	0	0	18	1	18	NaN	NaN	NaN
4	1871	BS1	NaN	gouldch01	31.0	NaN	31	31.0	0	...	0	0	0	1	1	NaN	NaN	NaN

	playerID	yearID	gameID	teamID	lgID	startingPos
573	borowha01	1945	NaN	NYA	AL	NaN
574	boudrlo01	1945	NaN	CLE	AL	NaN
575	casege01	1945	NaN	WS1	AL	NaN
576	chrisru01	1945	NaN	PHA	AL	NaN
577	cuccito01	1945	NaN	CHA	AL	NaN
578	ettenni01	1945	NaN	NYA	AL	NaN
579	ferreri01	1945	NaN	WS1	AL	NaN
580	ferrida01	1945	NaN	BOS	AL	NaN
581	greenha01	1945	NaN	DET	AL	NaN
582	grimeos01	1945	NaN	NYA	AL	NaN
583	gromest01	1945	NaN	CLE	AL	NaN
584	hayesfr01	1945	NaN	CLE	AL	NaN
585	johnsbo01	1945	NaN	BOS	AL	NaN
586	krameja01	1945	NaN	SLA	AL	NaN
587	leeth01	1945	NaN	CHA	AL	NaN
588	leonadu02	1945	NaN	WS1	AL	NaN
589	mayoed01	1945	NaN	DET	AL	NaN
590	mcquige02	1945	NaN	SLA	AL	NaN
591	moseswa01	1945	NaN	CHA	AL	NaN
592	newhoha01	1945	NaN	DET	AL	NaN
593	reynoal01	1945	NaN	CLE	AL	NaN
594	stephve01	1945	NaN	SLA	AL	NaN
595	stirnsn01	1945	NaN	NYA	AL	NaN
596	treshmi01	1945	NaN	CHA	AL	NaN
597	barrere01	1945	NaN	SLN	NL	NaN
598	cavarph01	1945	NaN	CHN	NL	NaN
599	coopemo01	1945	NaN	BSN	NL	NaN
600	elliobo01	1945	NaN	PIT	NL	NaN
601	greggha01	1945	NaN	BRO	NL	NaN
602	hackst01	1945	NaN	CHN	NL	NaN
603	holmeto01	1945	NaN	BSN	NL	NaN
604	johnsdo02	1945	NaN	CHN	NL	NaN
605	kurowwh01	1945	NaN	SLN	NL	NaN
606	lombaer01	1945	NaN	NY1	NL	NaN
607	marioma01	1945	NaN	SLN	NL	NaN
608	masiph01	1945	NaN	BSN	NL	NaN
609	mccorfr01	1945	NaN	CIN	NL	NaN
610	mungova01	1945	NaN	NY1	NL	NaN
611	nichobi01	1945	NaN	CHN	NL	NaN
612	odeake01	1945	NaN	SLN	NL	NaN
613	ottme01	1945	NaN	NY1	NL	NaN
614	pafkoan01	1945	NaN	CHN	NL	NaN
615	passecl01	1945	NaN	CHN	NL	NaN
616	roepr01	1945	NaN	PIT	NL	NaN
617	rosengo01	1945	NaN	BRO	NL	NaN
618	sewelri01	1945	NaN	PIT	NL	NaN
619	verbaem01	1945	NaN	SLN	NL	NaN
620	walkedi02	1945	NaN	BRO	NL	NaN
621	wyseha01	1945	NaN	CHN	NL	NaN

	yearID	teamID	lgID	playerID	G_all	GS	G_batting	G_defense	G_p	G_c	...	G_lf	G_cf	G_rf	G_of	G_dh	G_ph	G_pr
732	1875	WS6	NaN	stearbi01	21.0	NaN	21	21.0	17	0	...	2	5	0	7	NaN	NaN	NaN
733	1875	WS6	NaN	stevero01	1.0	NaN	1	1.0	0	0	...	0	0	1	1	NaN	NaN	NaN
734	1875	WS6	NaN	terry01	6.0	NaN	6	6.0	0	0	...	0	1	2	3	NaN	NaN	NaN
735	1875	WS6	NaN	thompan01	11.0	NaN	11	11.0	0	11	...	0	0	1	1	NaN	NaN	NaN
736	1875	WS6	NaN	withech01	1.0	NaN	1	1.0	1	0	...	0	0	0	0	NaN	NaN	NaN

	yearID	teamID	lgID	playerID	G_all	GS	G_batting	G_defense	G_p	...	G_lf	G_rf	G_of	G_dh	G_ph
96757	2013	ARI	NL	kenneia01	NaN	NaN	22	NaN	21	...	0	0	0	0.0	1.0
96758	2013	ARI	NL	kubelja01	NaN	NaN	89	NaN	0	...	53	5	57	3.0	30.0
96759	2013	ARI	NL	langwma01	NaN	NaN	8	NaN	8	...	0	0	0	0.0	0.0
96772	2013	ARI	NL	reynoma01	NaN	NaN	30	NaN	30	...	0	0	0	0.0	0.0
96779	2013	ARI	NL	thatcjo01	NaN	NaN	22	NaN	22	...	0	0	0	0.0	0.0

	playerID	yearID	bats	throws
24990	alexada01	1933	R	R
24991	almadme01	1933	L	L
24992	andreiv01	1933	R	R
24993	brownll01	1933	L	L
24994	cookedu01	1933	L	R

	players	allStars	percentAllStar
count	51.000000	51.000000	51.000000
mean	323.607843	28.980392	9.294504
std	415.579070	44.908124	4.281587
min	11.000000	1.000000	1.282051
25%	66.000000	5.000000	6.416137
50%	200.000000	17.000000	8.205128
75%	358.500000	35.500000	11.889098
max	2160.000000	295.000000	20.689655

bats	B	L	R
count	81.000000	83.000000	83.000000
mean	0.077125	0.294898	0.629836
std	0.037360	0.048871	0.045201
min	0.018868	0.189655	0.529412
25%	NaN	0.264428	0.599065
50%	NaN	0.289474	0.633803
75%	NaN	0.327200	0.660189
max	0.145161	0.396226	0.740741

bats	B	L	R
count	83.000000	83.000000	83.000000
mean	0.069939	0.286516	0.643545
std	0.026782	0.014535	0.023129
min	0.028571	0.253397	0.599034
25%	0.041688	0.276125	0.627119
50%	0.074521	0.284424	0.644461
75%	0.091829	0.294534	0.658926
max	0.109576	0.323993	0.705036

throws	L	R
count	83.000000	83.000000
mean	0.184127	0.815873
std	0.040392	0.040392
min	0.075472	0.721311
25%	0.164057	0.796610
50%	0.180328	0.819672
75%	0.203390	0.835943
max	0.278689	0.924528

throws	L	R
count	83.000000	83.000000
mean	0.202861	0.797139
std	0.019360	0.019360
min	0.157168	0.759553
25%	0.189421	0.781822
50%	0.204904	0.795096
75%	0.218178	0.810579
max	0.240447	0.842832