Predicting the Greek Elections: “Live” Evaluation

The table below presents the Twitter-based predictions along with the last week’s polls and the exit polls. The table will be updated once the actual results come out.

There were two exit polls (see “EP#1” and “EP#2” in the table) released at 19:00. Both of them provided the voting shares for every party as a range of values {minimum, maximum}. I considered their average as their actual “prediction”, making sure that all the adjusted voting shares sum up to 100.

For the case of the polls (see “P#1-8” in the table), I replaced all “undecided”, “blank” and “NA” voters’ voting shares and assigned them proportionally to the eight major political parties, as well as the “Others” (minor parties), again making sure that they sum up to 100. The Twitter-based method that I used is denoted as “TB” (“Twitter-Based”). “MAE” refers to the Mean Absolute Error of every method/poll. The results presented at the moment (00:00, UK time) are based on the 88.83% of the total votes.

Results TB EP#1 EP#2 P#1 P#2 P#3 P#4 P#5 P#6 P#7 P#8
SYRIZA

36.34

37.16

37.69

37.78

35.06

37.08

37.84

34.47

36.01

36.99

34.83

37.08

ND

27.85

28.31

25.13

25.69

30.34

31.40

25.44

29.19

30.66

29.57

30.34

31.40

Potami

6.02

5.90

7.24

7.56

7.08

6.49

6.39

7.05

7.18

5.54

6.74

6.49

XA

6.30

6.42

7.24

7.05

5.06

5.33

8.52

5.62

5.35

5.65

6.18

5.33

KKE

5.48

5.45

5.23

6.05

6.18

4.75

5.39

5.95

4.87

5.43

5.62

4.75

PASOK

4.69

5.10

4.72

5.04

4.49

4.63

4.26

4.74

4.01

5.54

6.18

4.63

ANEL

4.71

3.65

4.02

3.53

3.48

3.48

2.76

4.19

4.26

3.88

3.37

3.48

KIDISO

2.45

2.86

2.71

2.52

2.47

2.32

2.13

2.75

1.70

3.21

2.81

2.32

Others

6.16

5.15

6.03

4.79

5.84

4.52

7.27

6.06

5.96

4.21

3.93

4.52

MAE

0.00

0.49

0.84

1.05

0.95

1.06

1.16

0.71

0.88

0.88

1.15

1.06

The opinion polls were aggregated from here.

Advertisements

Predicting (?) the 2015 Greek Elections

Before proceeding, it should be noted that the predictions presented below should not be considered as valid estimations of the election results, but as part of an ongoing research. The reason they are published here is for providing evidence that the method was developed and applied before the elections and hence is not biased towards the actual results.

 

In our previous work we have exploited the potential of Twitter data to forecast the 2014 EU elections for three different countries. In the current research, we focus on the Greek elections, which have just begun and are expected to end at 19:00 (local time).

Since January, 9th we have been aggregating tweets written in the Greek language containing a political party’s name, its abbreviation or some common mispells. We have also aggregated opinion polls conducted during the same period. We extracted several features out of the aggregated data (Twitter-based and poll-based) and treated our task as a time-series forecasting problem.

The predictions of our method are summarised below. While several improvements to our past approach have been made (including a more appropriate sentiment analysis method and the use of Twitter-users’ weights), the training period was considerably shorter (16 days compared to 48 in the EU elections). Thus, the predictions should be read with caution. We plan to compare our model to all polls conducted during our processing time, as well as to the 19:00 Exit Polls.

 

Party Voting Share (%) Number of Seats
SYRIZA 37.16 151
New Democracy 28.31 77
Golden Dawn 6.42 17
To Potami 5.90 16
KKE 5.45 15
PASOK 5.10 14
ANEL 3.65 10
KIDISO 2.86 0
Others (total) 5.15

It should be noted that in order for a party to enter the parliament, a minimum of 3% of the total votes is needed, whie in order for a political party to form a government by its own, a minimum of 151 (out of 300) seats is needed.