cond() function

[Federico Nutarelli]

Hi all!

I have the following dilemma. I would like to create variables such that:

1. if a local takes on a certain name (i.e. price) the variable is created, otherwise it is not.
2. if a local takes on a certain name (i.e. price) the variable takes on a value otherwise another one.

I am doing this in order to perform regressions inside a loop. Specifically. as the code below suggests, I would like to perform regressions on "sales, stdunits and price". In particular having the local:

local expl "sales stdunits price"

in this step,

xtdata tasso_crescita_`k'_prod log_sales_lag `rec'_lag1 `rec'_lag2 `rec'_lag3 dumYear_2005 dumYear_2006 dumYear_2007 dumYear_2008 dumYear_2009 dumYear_2010 dumYear_2011 dumYear_2012 dumYear_2013 dumYear_2014 dumYear_2015 mean_gr_rate_atc2 ageprodcat4 ageprodcat3 ageprodcat2 ageprodcat1 recalls_`k' newmolfirm newmolmarket, fe clear

reg tasso_crescita_`k'_prod log_`k'_lag mean_gr_rate_atc2 `rec'_lag1 `rec'_lag2 `rec'_lag3 mean_gr_rate_atc2 recalls_`k' ageprodcat1 ageprodcat2 ageprodcat3 ageprodcat4 newmolfirm newmolmarket dumYear*

estimates store reg_product_`rec'_`k'

I would like to perform for sales and stdunits (I replace `k' just with sales but this remains valid for stdunits too):

xtdata tasso_crescita_sales_prod log_sales_lag `rec'_lag1 `rec'_lag2 `rec'_lag3 dumYear_2005 dumYear_2006 dumYear_2007 dumYear_2008 dumYear_2009 dumYear_2010 dumYear_2011 dumYear_2012 dumYear_2013 dumYear_2014 dumYear_2015 mean_gr_rate_atc2 ageprodcat4 ageprodcat3 ageprodcat2 ageprodcat1 recalls_sales newmolfirm newmolmarket, fe clear

reg tasso_crescita_sales_prod log_sales_lag mean_gr_rate_atc2 `rec'_lag1 `rec'_lag2 `rec'_lag3 mean_gr_rate_atc2_sales recalls_sales  ageprodcat1 ageprodcat2 ageprodcat3 ageprodcat4 newmolfirm newmolmarket dumYear*

estimates store reg_product_`rec'_sales

and for "price

xtdata price log_sales_lag `rec'_lag1 `rec'_lag2 `rec'_lag3 dumYear_2005 dumYear_2006 dumYear_2007 dumYear_2008 dumYear_2009 dumYear_2010 dumYear_2011 dumYear_2012 dumYear_2013 dumYear_2014 dumYear_2015 mean_gr_rate_atc2 ageprodcat4 ageprodcat3 ageprodcat2 ageprodcat1  newmolfirm newmolmarket, fe clear

reg price log_sales_lag mean_gr_rate_atc2 `rec'_lag1 `rec'_lag2 `rec'_lag3 mean_gr_rate_atc2_price  ageprodcat1 ageprodcat2 ageprodcat3 ageprodcat4 newmolfirm newmolmarket dumYear*

 estimates store reg_product_`rec'_sales

(hence price instead of tasso_crescita_`k'_prod, no variable recalls_sales in the reg with price and mean_gr_rate_atc2_`k' defined as mean(price), by(idatc2 Year) for price and mean(tasso_crescita_`k'_prod), by(idatc2 Year)) for sales and stdunits).

I am running a loop through local expl "sales stdunits price". To discriminate the "price" and "sales, stdunits" regressions I am using "cond()" to define variables, but it does not seem to work (I am doing something wrong). Specifically:

foreach rec of local recall {
foreach k of local expl {

        *PRODUCT
        use "/Users/federiconutarelli/Desktop/PhD/Lavoro_Riccaboni/DATi_USA/Pharma_project_2019 /abnormal_values/DB_product_`rec'_aggregation_nomiss.dta", clear
        
        * STEP 3: generare AV_t = y_it - E(y_it):
        eststo clear
        xtset idpr Year
        rename (tot_sales) (sales)
        gen `k'_lag = L.tot_`k'
        egen tasso_crescita_`k'_prod = ln(`k' - `k'_lag)
        gen log_`k' = ln(`k')
        gen recalls_lagged = L.`rec'
        gen log_`k'_lagged = L.log_`k'
        egen recalls_`k' = cond(`k'=="price",., recalls_lagged*log_`k'_lagged)

        replace ageprod = ageprod + 1
        gen ageprod_cat = .
        replace ageprod_cat = 1 if ageproduct <= 5
        replace ageprod_cat = 2 if ageproduct >= 6 & ageproduct <= 10
        replace ageprod_cat = 3 if ageproduct >= 11 & ageproduct <= 30
        replace ageprod_cat = 4 if ageproduct >= 31 & ageproduct < .
        tab ageprod_cat*, gen(ageprodcat)

        *Creating average growth rate of sales in ATC2 market:
        egen mean_gr_price = mean(price), by(idatc2 Year)
        egen mean_gr_rateatc2_`k' = cond(`k'=="price",mean_gr_price, (mean(tasso_crescita_`k'_prod), by(idatc2 Year)))
        *save "/Users/federiconutarelli/Desktop/DB_product_`rec'_`k'_nomiss.dta", replace
        
        * ESTIMATES WITH RECALLS GENERALI:
        gen log_`k'_lag = L.log_`k'
        gen `rec'_lag1 = L.`rec'
        gen `rec'_lag2 = L2.`rec'
        gen `rec'_lag3 = L3.`rec'
        xi i.Year, pre(dum)
        
        * E(Y-it):
        local vart
        su ritiri, meanonly
        quietly forval j = 1/`r(max)' {
            local vart `vart' rescaled`j'
            }
        
        local keep `vart' ritiri tasso_crescita_`k'_prod idproduct dumYear_2005 dumYear_2006 dumYear_2007 dumYear_2008 dumYear_2009 dumYear_2010 dumYear_2011 dumYear_2012 dumYear_2013 dumYear_2014 dumYear_2015 log_`k'_lagged log_`k' `rec' `rec'_lag1 `rec'_lag2 `rec'_lag3 mean_gr_rate_atc2 recalls_`k' ageprodcat1 ageprodcat2 ageprodcat3 ageprodcat4 newmolfirm newmolmarket Year  
        ds
        local vars `r(varlist)'
        local tokeep : list vars & keep

        keep `tokeep'
        
        *HERE I WOULD LIKE TO DISCRIMINATE BETWEEN "price" REGRESSION AND "sales stdunits" REGRESSIONS!
        local keep_no_rescaled  tasso_crescita_`k'_prod idproduct dumYear_2005 dumYear_2006 dumYear_2007 dumYear_2008 dumYear_2009 dumYear_2010 dumYear_2011 dumYear_2012 dumYear_2013 dumYear_2014 dumYear_2015 log_`k'_lag log_`k' `rec'_lag1 `rec'_lag2 `rec'_lag3 mean_gr_rate_atc2 recalls_`k' ageprodcat1 ageprodcat2 ageprodcat3 ageprodcat4 newmolfirm newmolmarket Year  
        * Keeping the observations that are not missing:
        foreach v of var `keep_no_rescaled' {
            drop if missing(`v')
        }
        
        
        gen id_unico = _n
        egen y_`k' = cond(`k'=="price",price,tasso_crescita_`k'_prod)
        
        save "/Users/federiconutarelli/Desktop/PhD/Lavoro_Riccaboni/DATi_USA/Pharma_project_2019 /abnormal_values/DB_product_`rec'_`k'_nomiss.dta", replace

        xtdata y_`k' log_sales_lag `rec'_lag1 `rec'_lag2 `rec'_lag3 dumYear_2005 dumYear_2006 dumYear_2007 dumYear_2008 dumYear_2009 dumYear_2010 dumYear_2011 dumYear_2012 dumYear_2013 dumYear_2014 dumYear_2015 mean_gr_rateatc2_`k' ageprodcat4 ageprodcat3 ageprodcat2 ageprodcat1 recalls_`k' newmolfirm newmolmarket, fe clear
        reg y`k' log_`k'_lag mean_gr_rate_atc2 `rec'_lag1 `rec'_lag2 `rec'_lag3 mean_gr_rateatc2_`k' recalls_`k' ageprodcat1 ageprodcat2 ageprodcat3 ageprodcat4 newmolfirm newmolmarket dumYear*
        estimates store reg_product_`rec'_`k'
    
        clear
    }
}

Am I doing something wrong? (Specifically in the steps of variables definition, e.g.

egen mean_gr_price = mean(price), by(idatc2 Year)
egen mean_gr_rateatc2_`k' = cond(`k'=="price",mean_gr_price, (mean(tasso_crescita_`k'_prod), by(idatc2 Year)))

or

egen recalls_`k' = cond(`k'=="price",., recalls_lagged*log_`k'_lagged)

)


Thank you very much!

Federico

[Sven-Kristjan Bormann]

The first obvious problem is that "cond()" does not work with "egen" but only with "gen". The second problem is that "cond()" does not accept a string comparison as its first argument. See the help file under "help cond()".
Therefore, I would replace "cond()" with another call.

gen recalls_`k' = recalls_lagged*log_`k'_lagged  if  "`k'" !="price"

If I am not completely mistaken then that should give the desired result.
You can replace the other calls to "cond()" with similar calls to "gen" and "replace".

[daniel klein]

The second problem is that "cond()" does not accept a string comparison as its first argument.

Not so!

. display cond("foo" == "foo", "bar", "foobar")
bar

The problem is that

`k'=="price"

should be

"`k'"=="price"

note the use of double-quotes.

Best
Daniel

[Federico Nutarelli]

Thank you both for the reply!

Appreciated a lot!

Daniel, indeed I made what you suggested, just, being price actually a variable in the code I put directly

`k'==price

So basically I removed double quotes also to price.
Do you think it's the same?

Thank you again!

[Nick Cox]

There was too much code in #1 to be tempting, but I have to point out that egen is compatible with cond() in so far as cond() can be part of an expression and many egen functions allows expressions as arguments. .

Examples in https://www.stata-journal.com/articl...article=dm0055 and indeed in many other places.

[Federico Nutarelli]

Thank you Nick. I will go and check out since both egen and cond() ar very much widespread in Stata!

[Sven-Kristjan Bormann]

daniel klein and Nick Cox Thanks for pointing out, where my answer was incorrect. I learned something from your corrections. Next time I check my answer more carefully before posting.