Elections project

Defining the files directory and paths

This is important for the Importing phase to loop on them when we need to access all the files and get all the names.

Phase : Importation

Import the files with pandas in a dictionary to loop on in the analysis

Phase : Create a preperation function

We Have to prepare the data in ‘dfs’ to get the total votes for each list,

# this function takes the dataframe sums row wise for each candidate 
# and return list name and the total votes
def prep(df):
    # act on the data frame and process it 
    df_0 = pd.concat([pd.DataFrame({'sumv':np.sum(df, axis=1)}), df['list']], axis = 1)
    return df_0

Phase: Define and prepare Hare quota arguments

Hare quota is the number of all votes in given city divided by the number of seats for that city. Based on the HQ we will create a table where we have:

• electoral quota for each list: Q = 1 means list gets 1 seat etc.
• seats collected fully by the votes quota, quota seats: QS.
• remains R from the quota: votes who didn’t got any seats to the list.
• percentage P of the votes of the list from all votes.

def hare(df, s):
    #total votes:
    ts = np.sum(df.sumv, axis = 0)
    #hare quota:
    hq = np.round(ts/s,decimals=3)
    #hare quota per list
    df['q'] = df.sumv/hq
    #quota seats
    df['qseats'] = np.fix(df.q)
    #remains
    df['r'] = df.q - df.qseats
    #percentage
    df['p'] = df.sumv/ts
    #sort the values with the highest remains first
    df = df.sort_values('r', ascending = False)
    return df

This is the result in Sousse:

py$tmp %>% kable

	sumv	list	q	qseats	r	p
14	102604	قائمة حركة نداء تونس	4.9290930	4	0.9290930	0.4929093
9	12360	قائمة حزب آفاق تونس	0.5937740	0	0.5937740	0.0593774
15	50820	قائمة حزب حركة النهضة	2.4413912	2	0.4413912	0.2441391
41	8626	قائمة حزب المبادرة	0.4143928	0	0.4143928	0.0414393
33	5502	قائمة الجبهة الشعبية	0.2643159	0	0.2643159	0.0264316

Phase : Prepare and perform computations for seats allocation

In this phase we will implement the largest remains allocation method for a given dataset. We will also give the opportunity to assign a minimum percentage of representation for lists to be accorded the remained seats. For each list sorted with regard to it’s remains, and satisfying the condition on the percentage, we add a seat, until all lists are given one, if more seats are still not allocated we iterate again with the same order until we have none, or until the first condition is met and we repeat.

def seats(df, s, p):
    df = df[df.p > p].reset_index()
    rs = np.int32(s - np.sum(df.qseats))
    for i in range(rs):
        df.qseats[i] = df.qseats[i] + 1
    return df[df.qseats > 0]

Phase : Combine all processing and computations

For future ease of use, testing, and debugging, it is convenient to create a function that combine all of the above. Let’s call it results().

def results(df, s, p):
    return seats(hare(prep(df), s), s, p)

here are the results in sousse

	index	sumv	list	q	qseats	r	p
0	14	102604	قائمة حركة نداء تونس	4.9290930	5	0.9290930	0.4929093
1	9	12360	قائمة حزب آفاق تونس	0.5937740	1	0.5937740	0.0593774
2	15	50820	قائمة حزب حركة النهضة	2.4413912	3	0.4413912	0.2441391
3	41	8626	قائمة حزب المبادرة	0.4143928	1	0.4143928	0.0414393

Phase : Get data on seats for each region

Using data from the wikipedia article on regional dispatching of parliamentary seats in Tunisia,

link.

Ofcourse we will not be needing all the page, only the table, the region, and seats associated.

الموقع	الدائرة الانتخابية	الأماكن	المقاعد
تونس (199 مقعد)	أريانة	ولاية أريانة	8
تونس (199 مقعد)	باجة	ولاية باجة	6
تونس (199 مقعد)	بن عروس	ولاية بن عروس	10
تونس (199 مقعد)	بنزرت	ولاية بنزرت	9
تونس (199 مقعد)	قابس	ولاية قابس	7

Because of the confusion between arabic letters and latin letters python orders them in a different manner so we needed to reorder them again in a consistent manner between arabic and latin letters.

seats	gov
8	ariana
6	beja
10	ben_arous
9	bizerte
7	gabes
7	gafsa
8	jendouba
9	kairouan
8	kasserine
5	kebili
7	mannouba
6	kef
8	mahdia
9	mednine
9	monastir
7	nabeul1
6	nabeul2
7	sfax1
9	sfax2
8	sidibouzid
6	siliana
10	sousse
4	tataouine
4	tozeur
9	tunis1
8	tunis2
5	zaghouan

Phase : Constructing the results

We will now use the table above, to loop for each region, it’s associated dataset in ‘dfs’, pass it to the functions one by one , using also the corresponding number of seats for each specified in the column seats.

We will associate the results into a dictionary we will call it ‘fr’, for final results, each key will be looped on as the name of the data set, and each value will be resulted dataset from the results function of the looped upon dfs dictionary.

In order to do this iteration, We will need to reorder the arrangement of names on which we created the ‘dfs’ dictionary, therefore, we need to reimport the data again in a proper manner.

Phase : Cleaning the names of the lists.

In this section I discovered That some names in arabic have been written in a terrible manner, this could create inconsistent results later on in the plotting, therefore we need to make the names of the lists that went to the parliament are clean.

0
قائمة حركة نداء تونس
قائمة حزب حركة النهضة
قائمة حزب الاتحاد الوطني الحر
قائمة الجبهة الشعبية
قائمة حزب التيار الديمقراطي
قائمة حزب التحالف
قائمة حزب المؤتمر من أجل الجمهورية
القائمة المستقلة الإقلاع
قائمة حزب صوت الفلاحين
قائمة تيار المحبة
قائمة حركة الشعب
قائمة حزب آفاق تونس
قائمة الجبهة الوطنية للإنقاذ
قائمة الوفاء لمشروع الشهيد
قائمة حزب المبادرة
قائمة المجد الجريد

We also needed to clean up some issues in the lists names like spaces, underscores, and more confusion because of arabic letters.

The results : a simple graphic

I only used mainland data here we haven’t integrated abroad citizens votes

tmp <- py$win
tmp %>% 
  mutate(list = fct_reorder(list, seats)) %>%
  ggplot(aes(x= seats, y = list)) +
  geom_col(fill = "sky blue") +
  geom_text(aes(label = tmp$seats), hjust = -.2, color = "grey") +
  theme_light()

Summary and conclusion:

I have used both python and R in order to achieve this, technically using them both is the ideal. The technical issues I faced during this little project, which I will keep updating and improving, are mainly related to the bad data quality that the ISIE is sharing with the public, the problem is that this is official legal and approved-by-the-court public data, the least that (the institution with the biggest budget in the entire country) can do is provide data with quality and clear standards. Putting all of this aside, I will be developing a shiny app to further demonstrate the effect of the percentage quota on the results really shortly. This was fun, looking forward for your feedback.

TO BE CONTINUED.