BRF GJP Trial Poll-1: How many seats will NDA win?

chaanakya · Post by **chaanakya** » 17 Apr 2014 22:48

Chaanakya 0 0 0 0 0 100

member_28108 · Post by **member_28108** » 17 Apr 2014 22:53

0 0 0 30 30 30 10

KLP Dubey · Post by **KLP Dubey** » 17 Apr 2014 23:14

0 0 0 0 15 35 50

Kanson · Post by **Kanson** » 17 Apr 2014 23:29

340 plus

chaanakya · Post by **chaanakya** » 17 Apr 2014 23:35

Kanson wrote:340 plus

well there is only 300+ as last choice. It could be more than 2/3rd as well.

Kanson · Post by **Kanson** » 17 Apr 2014 23:56

Of course, I understand the limitations of this trial. I'm trying to be more meaningful that simply stating 300 plus.
I'm starting to believe that many pollsters who were busy in micro analyzing individual seats never lifted their head to see the Tsunami coming.

rajithn · Post by **rajithn** » 18 Apr 2014 00:08

0,0,0,0,10%,10%,80%

ramana · Post by **ramana** » 18 Apr 2014 03:38

Thread locked for data processing.
Saral let me know in GJP thread for unlock.

ramana · Post by **ramana** » 18 Apr 2014 05:24

incremented the vote for this two weeks by+1. Will do the same after next round ie 4/28

RamaY · Post by **RamaY** » 18 Apr 2014 05:33

RamaY Week3

0 0 30 0 0 0 70

sivab · Post by **sivab** » 18 Apr 2014 05:35

ramana wrote:incremented the vote for this two weeks by+1. Will do the same after next round ie 4/28

Next vote is on 4/30.
http://www.thehindu.com/news/national/a ... theme=true

Saral · Post by **Saral** » 18 Apr 2014 05:43

Week 2 NDA

Table 1 is sorted in order of mean estimates (ESTIMATE) from each of the 33 distributions (rows). I use the center of each range as the weighting factor for each frequency (so 230.5 for the range 221 to 240). I used 180 and 320 as weights for the end ranges.

Table 2 is sorted by CONFIDENCE inferred from each distribution and is something made up for this collation. A perfectly uniform distribution indicates zero confidence (extreme hedging). At the other extreme, a confidence of 100 is assigned to people who put everything in 1 bin (zero hedging). Calculation: Each column in a row has an entropy term -p*log(p). Sum these up for each row and take the exponential of the sum. Subtract from 7, divide by 6, and multiply by 100 to map the score to the range 0 to 100.

ESTIMATE has mean=260, median=258, and SD=27. CONFIDENCE has mean=54, median=55, and SD=23. Consistent with literature, extreme estimates tend to be more confident as well. The correlation between deviation of ESTIMATE from the mean (of 260) and the CONFIDENCE score is 0.50. If we weigh each ESTIMATE by the CONFIDENCE, then the weighted estimate is 265.

Although the set of judges isn't the same, the median NDA estimate has increased to 258 (week 2) from 250 (week 1) even as the mean remains stable. The variation of estimates has declined (SD declined to 27 in week 2 from 33 in week 1), indicating more consensus across judges about the median estimate. The median confidence has gone up slightly but not by much (52 to 55).

Table 1

Code: Select all

        JINGO      NDA180 NDA210 NDA230 NDA250 NDA270 NDA290 NDA320    ESTIMATE     CONFIDENCE
       
      Santosh        30     35     25     10      0      0      0         210            55
    Shakimaan         0     40     40     10     10      0      0         228            62
         saip        10     10     60     10     10      0      0         229            60
       RahulM         5     30     40     15      5      5      0         230            45
       merlin         5     35     30     15     10      5      0         231            39
       Marten         5     30     30     20     10      5      0         233            36
         SRoy         5      5     40     50      0      0      0         237            71
   brihaspati         3     30     30     15     10     10      2         238            30
   HariSeldon         0     20     20     30     30      0      0         244            51
      matrimc         5     10     15     45     20      5      0         246            42
  MukeshKumar         5     15     25     20     20     15      0         246            25
       krisna         0      5     40     40      5      5      5         247            54
     rvishwak         0      0     25     65     10      0      0         248            77
        Klaus        30      4      4     30      0      0     32         249            53
         kish         2     10     25     25     25     13      0         250            33
        Denis         0      0     18     54     18     10      0         254            62
     Supratik         5     10     10     20     30     20      5         258            19
 vijaykarthik         0      0     12     38     38     12      0         260            59
   MortWalker         0     10     15     20     25     20     10         263            22
        disha         0      0      0     40     40     20      0         266            69
        Patni         0      0     10     30     40     15      5         266            50
Panduranghari         0      0      0     40     40     15      5         268            63
   mahadevbhu         5      5     10     10     20     40     10         270            27
         SaiK         0      0      0      0    100      0      0         270           100
   kapilrdave        20      0      0      5     20     25     30         271            42
       Khatri         5     10     10     10     15     25     25         271            14
        Saral         0      0      0     25     45     30      0         272            68
prasannasimha         0      0      0     30     30     30     10         275            55
        RamaY         0      0      0     10     20     30     40         294            57
     KLPDubey         0      0      0      0     15     35     50         302            71
      rajithn         0      0      0      0     10     10     80         312            85
       Kanson         0      0      0      0      0      0    100         320           100
    chaanakya         0      0      0      0      0      0    100         320           100

Table 2

Code: Select all

         JINGO     NDA180 NDA210 NDA230 NDA250 NDA270 NDA290 NDA320    ESTIMATE    CONFIDENCE
         
        Khatri         5     10     10     10     15     25     25         271            14
      Supratik         5     10     10     20     30     20      5         258            19
    MortWalker         0     10     15     20     25     20     10         263            22
   MukeshKumar         5     15     25     20     20     15      0         246            25
    mahadevbhu         5      5     10     10     20     40     10         270            27
    brihaspati         3     30     30     15     10     10      2         238            30
          kish         2     10     25     25     25     13      0         250            33
        Marten         5     30     30     20     10      5      0         233            36
        merlin         5     35     30     15     10      5      0         231            39
       matrimc         5     10     15     45     20      5      0         246            42
    kapilrdave        20      0      0      5     20     25     30         271            42
        RahulM         5     30     40     15      5      5      0         230            45
         Patni         0      0     10     30     40     15      5         266            50
    HariSeldon         0     20     20     30     30      0      0         244            51
         Klaus        30      4      4     30      0      0     32         249            53
        krisna         0      5     40     40      5      5      5         247            54
       Santosh        30     35     25     10      0      0      0         210            55
 prasannasimha         0      0      0     30     30     30     10         275            55
         RamaY         0      0      0     10     20     30     40         294            57
  vijaykarthik         0      0     12     38     38     12      0         260            59
          saip        10     10     60     10     10      0      0         229            60
         Denis         0      0     18     54     18     10      0         254            62
     Shakimaan         0     40     40     10     10      0      0         228            62
 Panduranghari         0      0      0     40     40     15      5         268            63
         Saral         0      0      0     25     45     30      0         272            68
         disha         0      0      0     40     40     20      0         266            69
          SRoy         5      5     40     50      0      0      0         237            71
      KLPDubey         0      0      0      0     15     35     50         302            71
      rvishwak         0      0     25     65     10      0      0         248            77
       rajithn         0      0      0      0     10     10     80         312            85
        Kanson         0      0      0      0      0      0    100         320           100
     chaanakya         0      0      0      0      0      0    100         320           100
          SaiK         0      0      0      0    100      0      0         270           100

krisna · Post by **krisna** » 18 Apr 2014 05:57

After todays elections- will continue the same as last phase.

cautious onlee.

ramana · Post by **ramana** » 18 Apr 2014 08:03

We are jingoes who want to win and not celebrate early!

saral, What kind of graph would make sense? Vijaykarthik please weigh in.

Saral · Post by **Saral** » 18 Apr 2014 08:33

Once May 16 arrives, we could plot median estimates across time, consensus between judges (SD of estimate) across time, confidence across time, and finally median accuracy over time. Accuracy could be the squared deviation of each judgment (probability distribution) from the actual value (which could be 238 or 308 or ~~178~~). We should have 6 sets of judgments before this is over.

More than plots, what would be more interesting is asking why people make the judgments they do. In particular, I am curious about judgments that are spread over a wide range. This is different from saying that either > 250 or < 200 depending on two kinds of scenarios (there was one judgment that has this pattern). In my judgments I try to narrow down the range so that most of the numbers are in a few ranges; so if you look at my week 2 it has shifted up a little bit and also narrowed as well.. this is the sense I got from the chatter. Some folks have decided early on that > 300 is the only option that makes sense. So if it does turn out that way, they are the winners.

brihaspati · Post by **brihaspati** » 18 Apr 2014 09:39

3rd week : 0 10 35 35 10 5 0

Yagnasri · Post by **Yagnasri** » 18 Apr 2014 10:09

NDA minus telugus will be 220 to 230. Even in a wave other side will win seats. No wave will help if party is weak in some locality. Some individual candidates are simply impossible or too hard to defeat. Mafia and it's b teams have at least 50 plus such seats. Plus nda and regional parties are not strong in few seats even in their strongest states. Mafia plus got 140 plus in 2004. Now it will be around 120. There are at least 100 seats wherein peacefuls can effect the out come. Now most of these seats are with regional parties many of whom may not like nda. So near 250 is must.

By the way I am doing work for Shiva Sena as my locality seat alloted to it.

HARA MAHADEV.

Denis · Post by **Denis** » 18 Apr 2014 15:20

3rd week : 0 0 15 15 60 10 0

chaanakya · Post by **chaanakya** » 18 Apr 2014 15:33

Chaanakya 0 0 0 0 0 100

ramana · Post by **ramana** » 18 Apr 2014 20:17

Folks just to make it easy for saral, when there is no change or same as before please post the same number again. A request.

ramana

member_28108 · Post by **member_28108** » 18 Apr 2014 20:33

No change

vijaykarthik · Post by **vijaykarthik** » 18 Apr 2014 20:52

vijaykarthik 0 0 12 38 38 12 0 [No change, basically]

BTW, I forgot to mention it earlier. the same qn is also there on inklingmarkets. I put it after we spoke about it. Have a shot if you like:
http://home.inklingmarkets.com/markets/61049

RamaY · Post by **RamaY** » 18 Apr 2014 21:02

So it will not be missed

RamaY Week3: 0 0 30 0 0 0 70

vijaykarthik · Post by **vijaykarthik** » 18 Apr 2014 21:08

@RamaY -- is there a secret behind your numbers. Curious.

Its either about 240 or >300. Why do you think there is a possibility of nothing in between?

Supratik · Post by **Supratik** » 18 Apr 2014 22:35

5 10 10 20 30 20 5

devesh · Post by **devesh** » 19 Apr 2014 01:21

~~devesh 5 30 50 15 0 0 0~~

Revision: didn't realize this for NDA coalition, and not just BJP. Sorry!

devesh week3: 0 10 20 30 40 0 0

panduranghari · Post by **panduranghari** » 19 Apr 2014 01:49

0 0 0 30 50 15 5

Mort Walker · Post by **Mort Walker** » 19 Apr 2014 03:07

Mort Walker Week 3: 0, 0, 15, 20, 35, 30, 0

rajithn · Post by **rajithn** » 19 Apr 2014 03:29

rajithn week 3: 0%, 0%, 0%, 0%, 10%, 10%, 80%

RamaY · Post by **RamaY** » 19 Apr 2014 06:46

vijaykarthik wrote:@RamaY -- is there a secret behind your numbers. Curious.

Its either about 240 or >300. Why do you think there is a possibility of nothing in between?

Me think either BJP will get <200 or >272. It's either Modi wave or no Modi wave.

brihaspati · Post by **brihaspati** » 19 Apr 2014 08:35

Can I change my dist given earlier for 3rd to 0 0 15 70 15 0 0 ?
I have also strong inclination to have the final tally beginning with 2 and ending in a 8. Not sure about the middle digit.

ramana · Post by **ramana** » 19 Apr 2014 09:05

Saral · Post by **Saral** » 19 Apr 2014 18:23

All revisions valid till thread is locked next week

Rahul Mehta · Post by **Rahul Mehta** » 19 Apr 2014 18:45

Less than 200 - 0%
Between 200-220 - 5%
Between 221-240 - 5%
Between 241-260 - 10%
Between 261-280 - 30%
between 281-300 - 30%
or greater 300 -10%

UP is landslide, over 2/3 seats in UP. And of course, Gujarat, some 23/26 .

ramana · Post by **ramana** » 19 Apr 2014 20:34

I think we are now seeing the predictions taking into account the Congress systemic collapse. Early on it was based on linear extrapolation of past elections. Despite candidate selection mis-steps to accommodate non-performers who can sabotage, we are seeing a Modi wave.

SwamyG · Post by **SwamyG** » 19 Apr 2014 21:01

15 5 5 10 20 20 5 - Just BJP, not including NDA.

devesh · Post by **devesh** » 19 Apr 2014 21:15

Saral ji,

I'm reposting my revision here in case the Edit above is missed. it's the same as the above revision. not a "new" revision

devesh week3: 0 10 20 30 40 0 0

also, quick question: how are you calculating the confidence levels? are you assuming that the % number predicted is uniform for all data point within each bracket? for example, in my prediction above, I said 10% probability of NDA getting 210. so in you calculations, are you assigning 10% to all numbers between 180 and 210?

thanks!

Saral · Post by **Saral** » 20 Apr 2014 19:22

Saral 0 0 0 10 65 25 0

Devesh: Please see the table notes above table 1 for CONFIDENCE definition. This is something I have proposed (not taken from anywhere, so I do not currently have a citation for this particular usage) for this exercise. The 2 extremes are 0 confidence and 100 confidence. 0 means that all the 7 ranges are assigned the same percentage value (same thing as saying you don't have any information to differentiate the ranges) and 100 means that you put 100 inside one range. In-between values are calculated as per the note in the table. If you want to calculate my confidence for week3 paste this in google search which automatically uses its calculator and doesn't actually search:

Code: Select all

100*( 7 - exp(-(0.1*ln(0.1)+0.65*ln(0.65)+0.25*ln(0.25))))/6

this gives a value of 77.4 or 77 (after rounding) which is quite high. Reason being I am betting 65 on the 260 to 280 range and have left 4 ranges with zero. In earlier weeks, my estimates were more spread out. If for some reason, things go haywire a week from now, I might have to spread out and also reduce the modal range as well. But in poll prediction, things are supposed to get more certain and not less with incremental information.

P.S. I think its helpful from a post-mortem perspective, if people try to explain their rationale on why they are assigning numbers the way they are.

In my case, for week 3, its a case of trying to decide the modal range where the max prob lies. Based on various sources of info (NDTV poll, 5forty3, rumors on twitter) it seems that BJP alone should manage 230. So it seems that NDA should cross 272. What I did this week is simply change the numbers so that my mean is slightly more than 272. My estimates are not based on a state by state analysis etc. but more based on anchoring and adjustments.

Factors of concern to me: Reports that BJP will draw 0 in Kerala.. But in any case they weren't expecting more than 1 at the max. Maharashtra issues pertaining to voter rolls (todays news claims 60 lakh persons omitted.. not sure how such screw ups can happen).. Even in UP BJP vote share is around 36 according to some but given the nature of the contests (3 sided or more), maybe this is enough for a landslide of 50 or more out of 80. It would be interesting to hear from people whose modes are much less than 240. For those who are putting non-zero numbers for 300 and above, obviously you are postulating a wave that yields seats (and not just increased vote percentage).

RKumar · Post by **RKumar** » 20 Apr 2014 19:30

My first and final estimate

0 10 35 35 10 10 0

My reasons
1. Typical attitude, my 1-2 votes will not make a difference.
2. There is AAP effect.
3. Back stabbing, black mailing & Cheating by seculars
4. people are talking but not voting for NaMo

RKumar · Post by **RKumar** » 20 Apr 2014 19:37

BTW in poll, I voted for before election started or during first week
between 281-300

Based on the feedback from people I know, I am lowering the chances for NDA. It is shame but NaMo has to work harder. I will be more then happy if my first estimate still holds. That way smaller parties can't blackmail NaMo on important issues.

Next week is really important, I hope NaMo can keep the momentum as others have nothing to talk except NaMo bashing