clear	
	clear matrix
	set matsize 800
	set mem 500m
	cd your partition
	use [lanham pitching data]



	** back out hitter ABs
	*


	replace BAOpp = . if BAOpp == 9.99

	gen abf= H/BAOpp  

	drop if abf==.



	// sum data per pitcher



	collapse (sum) W L G GS CG SHO SV H ER HR BB SO  IBB WP HBP BK BFP GF R  GIDP IPouts abf, by(playerID yearID teamID)


	// IP

	gen IP=IPouts/3

// merege Lahman names to integrate 2024 data

	merge m:m playerID using  [lahman names],update
	drop _merge




/// BIP files


	merge m:m playerID yearID using 1930-79 BIP data,update

	
	drop _merge
	
	
		merge m:m playerID yearID using 1980-99 BIP data,update

	
	drop _merge
	
	
	
	merge m:m playerID yearID using 1912-29 BIP data ,update

	
	drop _merge
	
		
	
	
	
merge m:m playerID yearID using 2000-2024 BIP data, update replace
	
	drop _merge

	
	
// correct a possible glitch in Lahman team-name key for old retrosheet; I think this was fixed upstream but let's be sure
	
 replace teamID ="MIL" if teamID == "ML4" & year > 1997
 replace teamID ="ML4"  if teamID == "MIL" & year <1998
 replace teamID="LAA" if teamID =="ANA" & year >2004
	
/// merge bbref 2024 pitcher data	
	

merge m:m bbrefID yearID using bb_ref_2024_pitcher_patch,update
drop _merge
drop if yearID==.


/// merge bbref SH SF for obabip calculations
	merge m:m bbrefID yearID using "bb_ref_pitcher_sh_sf.dta",update 
	drop _merge
	drop if yearID==.

replace abf=(H+(IP*3)) if year ==2024


*** add knuckleballer identifier

gen kb=0  
gen kb_plus=0

merge m:m playerID using "knuckleballers", update replace

** kb is ever relied on knuckler
** kb_plus is principally relied for at least portion of career


drop _merge	


	
/// narrow to ip 100+
drop if IP<100
drop if ifpop==.

drop if SO ==.


/// abf for 2024

replace abf=(H+(IP*3)) if year ==2024


*** determine performance values as rates

gen k_pip=SO/IP
gen bb_pip=BB/IP
gen hr_pip=HR/IP
gen hbp_pip=HBP/IP
drop if IP <100
gen rapg=9*R/IP
gen obabip=(H-HR)/(abf-(SO+HR)+SF)
gen gb_pct= gb/(gb+airb)
gen ifgb_pip=ifgb/IP
gen ifpop_pip=ifpop/IP
gen airb_pip=airb/IP
gen ofair_pip=ofair/IP

// fipr scores

regress rapg k_pip bb_pip hr_pip hbp_pip [pweight=IP]
predict fipr 

	
// merge team fielding

gen teamid = teamID


 merge m:m  teamid yearID using "C:\Users\dmk38\Documents\x5\team_field_composite.dta"
 
 drop _merge


	
		
///variables for bip_rba


* Generate 7 run-environment-period models

bysort playerID (yearID): replace bip_comb_minus1 = bip_comb[_n-1] if playerID == playerID[_n-1]
bysort playerID (yearID): replace bip_comb_plus1 = bip_comb[_n+1] if playerID == playerID[_n-1]
gen period = .
replace period = 1912 if yearID < 1920
replace period = 1920 if yearID >= 1920 & yearID < 1942
replace period = 1942 if yearID >= 1942 & yearID < 1963  
replace period = 1963 if yearID >= 1963 & yearID < 1977
replace period = 1977 if yearID >= 1977 & yearID < 1993
replace period = 1993 if yearID >= 1993 & yearID < 2010
replace period = 2010 if yearID >= 2010


foreach p in 1912 1920 1942 1963 1977 1993 2010 {
   di "model period" `p'
   reg rapg fipr field_composite ofair_pip ifpop_pip [aweight=IP] if period==`p'
   
   gen beta_ofair_`p' = _b[ofair_pip] if period==`p'
   gen beta_ifpop_`p' = _b[ifpop_pip] if period==`p'
   
   mean ofair_pip ifpop_pip [aweight=IP] if period==`p'
   gen mean_ofair_`p' = e(b)[1,1] if period==`p'
   gen mean_ifpop_`p' = e(b)[1,2] if period==`p'
}


gen beta_ofair = .
gen beta_ifpop = .
gen mean_ofair = .
gen mean_ifpop = .
foreach p in 1912 1920 1942 1963 1977 1993 2010 {
   replace beta_ofair = beta_ofair_`p' if period==`p'
   replace beta_ifpop = beta_ifpop_`p' if period==`p'
   replace mean_ofair = mean_ofair_`p' if period==`p'
   replace mean_ifpop = mean_ifpop_`p' if period==`p'
   drop beta_ofair_`p' beta_ifpop_`p' mean_ofair_`p' mean_ifpop_`p'
}

// bip_rba

gen bip_rba=-1*(beta_ofair*(ofair_pip-mean_ofair))+beta_ifpop*(mean_ifpop-ifpop_pip)

// impact bip_rba on obabip

	gen obabip_suppress=.
	
foreach p in 1920 1942 1963 1977 1993 2010 {
  di as text "{yellow}period `p'{reset}"
   
   regress obabip field_composite bip_rba [aweight =IP] if period==`p'
   replace obabip_suppress=_b[bip_rba] if period==`p'
   
   
   
	   }




/// filter for IP; don't execute if want all at >= 100 IP

drop if IP < 154 & year <1961
drop if IP < 162 & year >1960



*** BIP_RBA reliability
	
*** BIP_RBA reliability
	
* Step 1: Sort the data by playerid and yearID
sort playerID yearID

* Step 2: Generate season -1 & season +1 variables
gen bip_rba_minus1 = .
gen bip_rba_plus1  = .

	
bysort playerID (yearID): replace bip_rba_minus1 = bip_rba[_n-1] if playerID == playerID[_n-1]
bysort playerID (yearID): replace bip_rba_plus1 = bip_rba[_n+1] if playerID == playerID[_n-1]



alpha (bip_rba_minus1 bip_rba_plus1 bip_rba)

** export player scores

drop if nameLast==""


egen r_bip_rba = rank(bip_rba)
gen sample_percentile= 100 * (r_bip_rba - 1) / (_N - 1)



export excel nameFirst nameLast yearID IP bip_rba  kb_plus sample_percentile  using "bip_rba_rps_periods2_plusknuckles.xls", firstrow(variables)   replace


*** likelihood of knuckleball ps being in 80th and 90th percentiles
gen p80=0
replace p80=1 if sample_percentile >=80 
logit p80 kb_plus [pweight =IP]
predict phat

su phat if kb_plus==1,d

gen p90=0
replace p90=1 if sample_percentile>=90
su phat if kb_plus==1

logit p90 kb_plus [pweight=IP]
predict phat2
sum phat2 if kb_plus==1

export excel nameFirst nameLast yearID bip_rba kb_plus sample_percentile obabip_suppress  using "bip_rba.xlsx", firstrow(variables)   replace