Thursday, July 17, 2014

Friday, April 11, 2014

Turkish Local Elections: Voting Behavior in Istanbul


Tactical Voting

Before and after the Turkish local elections held in March 2014, tactical voting has been discussed by the pundits in media.  Especially, compromising (useful vote) where a person votes for a party that has a higher chance of defeating her least favorite party instead of voting for the candidate she prefers the most. Even the candidates of AKP mentioned a collaboration between the opposition parties to defeat AKP's candidates in big cities such as İstanbul and Ankara. Another discussion point was Sırrı Süreyya Önder's potential withdrawal from the İstanbul mayor race to fortify Mustafa Sarıgül's (CHP) position as the sole opponent of the incumbent mayor Kadir Topbaş (AKP).

I tried to visualize the tactical voting behavior using the share of votes scored by each political party at the district level in İstanbul. Underlying assumption is that a voter is likely to choose her highest ranked party (candidate) for the district (ilçe) mayor, but comprise for the büyükşehir (city) race. All our conclusion rely on this assumption. 

Graph below shows the difference in the vote share between the büyükşehir and ilçe elections per party. First of all, we notice that HDP voters in most districts did not compromise. Secondly, Sırrı Süreyya Önder did not get a higher share than ilçe candidates showing that he did not steal votes from CHP following his increased popularity after the Gezi Parkı protests. Lastly, we see that MHP was the most negatively affected party by tactical voting. Though we can see that MHP voters comprised to both AKP and CHP.

Vote Share Difference between Büyükşehir and İlçe Mayor Elections in İstanbul





Affinity to a Party

As an additional analysis, I wanted to visualize the affinity of each district towards the major political parties, where distance is simply the vote share and a higher share indicates a closer tie. I actually wanted to include HDP as the fourth dimension; however, I could not figure out how to plot a tetrahedron in R. When/If I do, I will revise the graph. I decided to add HDP as the color layer. This works pretty nicely since a more common discussion involving HDP is crossing the 10% election threshold for the general elections in Turkey. 





How to: Ternary graphs are usually used to represent 3-dimensional information on a 2-dimensional surface when the data for an observation sums up to 1(or 100%) such as probabilities for three outcomes or results of a survey with three choices.  I used the ggtern package in R to produce the graph. Since the data points are densely located, using the geom_text() resulted in an illegible graph. R command is in the below form:


>ggtern(data=data,aes(CHP,AKP,MHP,col=HDP,label=ILCE))+geom_text(size=2.5)+theme(legend.position="bottom")+ggtitle("ISTANBUL ILCE BELEDIYE BASKANI OY ORANLARI")

Data input is a dataframe object as show below.



Finally, I used Inkscape to label the graph. It is a fairly straight forward tool to use, and there are tutorials on Youtube.

Wednesday, April 9, 2014

YILLARA GÖRE KARAYOLU UZUNLUĞU


Yerel seçimler öncesinde Tayyip Erdoğan AKP hükümetleri dönemindeki icratlara ve ilerlemelere örnek olarak yapılan bölünmüş yolların uzunluğunu gösterdi. Tabi yolların uzunluğu tek başına çok anlamlı değil; yolların cinsi de önemli. Viyadük ve tünellerin seyahat mesafesini kısaltması bir eksi değil artıdır. Kara Yolları Genel Müdürlüğü'den aldığım verileri görselleştirdim. (Burada ve burada)  Bölünmüş yol grafiklerinde son yıl 2012'dir.






AKP'nin iktidara 3 Kasım 2002'de geldiğini göz önüne alarak 2003'den sonrak' döneme bakarsak...




WSJ Battle of the Sexes Graph

                                  

Prof. Steve H. Hanke posted the graph above from WSJ on Twitter. It is hard to be precise with only 140 characters, but there is more to the story than the title "Battle of the Sexes" implies. If you read the article (here), wording of the BLS economist is very careful. It is not a choice for a sex over the other, but rather it is due to where the jobs were lost. If you look at the replies to Prof.Hanke's post, it appears that the general perceptions is: 

1) Higher participation of women to labor force 
2) Increasing share of women in the labor force

Let's look at the data!

As mentioned in the article, one factor is where the jobs were lost such as the construction sector , where the share of women is far smaller. Does anyone expect it to revert to the all time high? 


Let's look at women's share in the Total Nonfarm number and in the Private Service Sector number. Neither shows an increased share of women.


And the worst part, labor participation rate is at the lows for men and women... 






Tuesday, April 8, 2014

Yet Another Post Regarding Local (Municipal) Elections in Ankara and Istanbul

* Comments below or to @AKirayoglu are welcomed.

I had two previous posts (in Turkish) regarding the local government elections in Turkey, mostly focusing on the results and possible irregularities in Ankara. This post takes another look at the results in Ankara and Istanbul.

Data: Data was obtained from http://sts.chp.org.tr since the official results have not been publish by the YSK. As the results are being updated, data may have changed since I scrapped it. I excluded the votes cast in prisons. For replication purposes, you can use the links to download the data sets.


Ankara: http://bit.ly/1inPEBs
Istanbul: http://bit.ly/1kErTGG

Approach: Data set includes information such as number of registered voters, number of votes cast and number of invalid votes for buyuksehir (city) mayoral elections as well as ilce (district) mayoral elections at the ballot box level. It also includes the number of votes cast per political party for the buyuksehir mayoral elections.

Previous studies used various approaches to control for differences across districts and voting stations which may influence the share of invalid votes and the share of votes for a particular party simultaneously. Our assumption is that buyuksehir and ilce votes will be influenced by unobserved variables such as income, education or age at the ballot box level.

We will be using a 2-stage approach. First, we use the relationship between the buyuksehir and ilce elections and estimate the linear relationships below:

Buyuksehir % Invalid Votes =  a + b*(Ilce % Invalid Votes) + u
Buyuksehir Turnout = c + d*(Ilce Turnout) + v




For Ankara
Invalid Votes                
Model

Turnout
Model
(Intercept) 0.01*** 0.02***
(0.00) (0.00)

Invalid Votes Share District        
Election
0.75***
(0.01)
Turnout District Election 0.98***
(0.00)
R2 0.59 0.95
Adj. R2 0.59 0.95
Num. obs. 12234 12234
***p < 0.001, **p < 0.01, *p < 0.05


For Istanbul

Invalid Votes                
Model

Turnout

Model
(Intercept) 0.01*** 0.04***
(0.00) (0.00)

Invalid Votes Share District        
Election
0.68***
(0.00)
Turnout District Election 0.95***
(0.00)
R2 0.53 0.92
Adj. R2 0.53 0.92
Num. obs. 32164 32164
***p < 0.001, **p < 0.01, *p < 0.05

In the second stage, we regress the AKP vote share on the residuals from the first stage models and dummy variables for districts and voting stations. Standard errors are clustered. For Ankara, when we control at the ilce level coefficients for residuals from the first stage are positive and statistically significant for both turnout and invalid vote share models. The positive relationship remains, but the statistical significance disappears when we control at the voting station level. For Istanbul, invalid vote residuals have a positive and statistically significant relationship when controlled at both ilce and voting station level. However, turnout is not statistically significant. It is surprising to see that for both Ankara and Istanbul when the buyuksehir voting characteristics deviate from the predictions, we find statistically significant relationships where positive deviations indicate increase in AKP's vote share.



For Ankara Invalid Votes 
Model
Ilce Dummies 
Invalid Votes                   
Model
Alan Dummies
Turnout 
Model 
Ilce Dummies
         Turnout
         Model
        Alan Dummies


Invalid Votes Model        
Residuals

0.5097**
(0.1916)

0.0629
(0.0540)

Turnout Model Residuals

0.2338**
(0.08577)
 
     0.0945
 0.0760

***p < 0.001, **p < 0.01, *p < 0.05


For IstanbulInvalid Votes 
Model
Ilce Dummies 
Invalid Votes                   
Model
Alan Dummies
Turnout 
Model 
Ilce Dummies
         Turnout
         Model
        Alan Dummies


Invalid Votes Model        
Residuals

0.7234***
(0.1053)

0.1936***
(0.0352)

Turnout Model Residuals

0.03003
(0.09016)

   
-0.0157
(0.0475)
 

***p < 0.001, **p < 0.01, *p < 0.05






God, Corruption, GDP in One Graph

After Conrad Hackett shared the below data on Twitter. (More here) I was curious to see the correlation to a corruption index, and then realized perhaps income in the confounding factor.




Here is my graph:



Monday, April 7, 2014

Mahalli Seçimler - Daha Önceki Çalışmalar ile İlgili Bazı Görüşler


Daha önce yapılan çalışmalar hakkında bazı görüşler

 İlgi çeken çalışmalar @sandikbugları'nın twitter üzerinden paylaştığı veriyi kullanarak @emeyersson tarafından paylaşıldı: http://erikmeyersson.com/blog/ Bu çalışmalarda benim gözüme çarpan birkaç nokta oldu.

İlk olarak katılım oranı ve AKP oy oranı ilişkisi incelendiği zaman küçük olan sandıkların etkisinin göz önüne alınması gerektiği kanısındayım. Zira küçük sandıklarda görevli memurlar tarafından kullanılabilecek yasal ek oylar katılım oranına daha büyük etki edecektir. Mesela İstanbul ve Ankara karşılaştırıldığı takdirde İstanbul'un nüfusu daha yoğun olduğu için daha az küçük sandık olduğunu görüyorsuz. Bu da katılım oranı ve AKP oy oranı ilişkisini gösteren grafiklerin İstanbul ve Ankara arasında farklılık göstermesine neden oluyor.



İkinci nokta ise AKP-CHP arasındaki oy oranı farkı ve geçersiz oy oranını inceleyen ilişki ile alakalı. Erik'in fixed effect modelleri kullanarak yaptığı çalışmada ilçe ve alan bazında kontrol kullanıldığı takdirde geçersiz oyların etkisinin azaldığı görülmekte.

Kaynak: http://bit.ly/1hwVEtI


Bir diğer nokta ise ilçe seviyesinde kontrollerin yeterli olup olmadığı. Burada paylaşılan ilçe bazında yapılmış çalışmayı tekrarladım. Sonuçlar biraz farklı olsa da genel olarak aynı doğrultuda.




Peki residuallara baktığımızda ne görüyoruz. Residuallardaki gruplaşmalar ilçelerdeki FE modelinin doğru tanımlanmamış olduğunu gösteriyor olabilir mi?