Paper by Hans L Zetterberg published 1979 as a free booklet from Sifo AB, Vällingby. Also presented at and published in the transactions from a symposium on election polling arranged by Statistics Sweden: Rapport från SCBs forskarkonferens i statistisk metodik, Gimo herrgård, 16-18 maj 1979, Stockholm 1979, pp 12-54.

Sifos Election Poll and Its Deviations, 1967-1978

Sifos Election Poll
Pre-Election Polls and Election Outcome
Deviation from Trend (Margin of Error)
Deviation from Sampling Plan (Nonresponse)
Tables
Questionnaire and Ballot Papers

 

To Detailed Table of Contents

 

Translated from Swedish by Clare James
Click here for Swedish original
Svensk text

This report contains data on the Sifo election poll for the Swedish daily newspapers Dagens Nyheter (DN), GöteborgsPosten (GP) and Sydsvenska Dagbladet (SDS). It supersedes corresponding Sifo reports from  1970, 1973, 1975 and 1977.

The Sifo election poll is intended for political journalism. Its purpose is to give a quick guide to trends of public party-political sympathies under the influence of political developments. The journalistic interest should be emphasized, since it deviates from the scientific. Scientists can wait a year or more until a party trend is established and confirmed by numerous measurements, and sometimes also general elections. Journalists want to know about changes immediately. Election polls seek to satisfy the needs of journalists by approximating changes, as early as possible, what can only later be ascertained more precisely as a real change.

Figures on political parties in the Sifo election poll since its inception in March 1967 are shown on pages 1-3 of the table supplement. The figures are rounded to the nearest whole or half percentage point (except for small parties, whose totals are for which their total is so rounded). This rounding is carried out so as not to encourage decimal nitpicking in comments on the election poll.

Sifo's election poll is a monthly opinion survey in which the question posed is "What party do you think is the best today?" It is conducted within the framework of Sifo's continuous interview series, the so- called "weekly omnibus."den s k veckobussen (har du ett förslag?) After notification by mail and (usually) agreement by telephone, the interviews take place when the interviewers visit the selected respondents at home.

The timetable for the interviews is set one year in advance. Sifo delivers the results of its election polls to its customers, DN, GP, and SDS eight to ten days after the last interview date.

The first block of questions in the omnibus survey relate to consumer confidence and household finances. The second is the election poll itself. Other question blocks in an omnibus are unknown to the interviewees at the time when they reply to the questions of the election poll, and thus cannot affect the responses in the poll.

The questionnaire is reproduced on the last page of this report. See pages 37-38 for the cover of the questionnaire, available in advance by interviewers to book their interviews and attempts to contact interviewees, and page 39 for the questions of the election-poll questions and ballot-papers.

The replies concerning party support are placed by the interviewees in ballot envelopes and then in outer envelopes, and are unknown to the interviewers. The interviewers send the envelopes with the rest of the interview questionnaire to Sifo. After they are ticked off for field checking, the names are removed from the questionnaire (see the perforation on the first page of the cover to the questionnaire). The envelopes are opened later  when the questionnaires are coded. Thus, no employee at Sifo can connect information about a party on a particular questionnaire with a specific person. Moreover, both the interviewers and the rest of the staff are obliged to treat as confidential the facts concerning individuals that come to their knowledge through the interviews.

The purpose of the election poll is not to specify what the outcome of an election should be - which is impossible, since for the lion's share of the polls no election campaign has been currently underway - but to indicate the trends of the various party sympathies.

Pre-election polls and election outcome

In the final pre-election polls carried out since 1970 immediately before elections, the question posed has been changed to: "Which party do you intend to vote for in the parliamentary election?" The responses have not been rounded to the nearest whole or half percentage point. The reason for this has been to permit better comparisons with the election outcome. The results of these comparisons have been as follows:

 

Percentage distribution

No of inter-
views

Non
response
%

m

fp

c

s

vpk

1970
Sifo
Election


11.5
11.5


17.0
16.2

18.9
19.9

45.5
45.3

3.7
4.8

1398

11.6

1973
Sifo
Election

13.5
14.3

10.3
9.4

24.3
25.1

43.1
43.6


5.1
5.3

1140

14.8

1976
Sifo
Election

5.3
5.6

11.2
11.1

22.0
24.1


43.8
42.7

5.1
4.8

1013

15.7

m = Moderate Party
fp = Liberal Party
c = Center Party
s = Social Democratic Party
vpk = Left Party Communists

[The number of interviews conducted in 1973 was stated as 1,338 in the original. The correct figure is the one given above: 1,140.]

Sifo's pre-election polls are published before Election Day. The interviews are conducted with people who were polled after the previous election. The nonresponse rate can therefore be raised by the primary nonresponse that occurred in the source polls (pages 22-23).

A sound indication of the trend of party sympathies -what the election poll sets out to measure - is provided by a centered, sliding mean. We use a nine-month trend cycle without seasonal factors, with weights according to Henderson, to calculate the trend. An adequate trend figure for a given month can be obtained only six months later, when all the initial figures required for the sliding mean are available.

The party trends are shown on pages 4-6. On pages 7-8 the changes in party sympathies between 1967 and 1978 is tabulated.

It should never be forgotten that all sample surveys give approximate results only, not exact ones.

Note that Sifo does nor question the eldest people (currently over 70) entitled to vote, nor, eligible voters outside Sweden. Note also that not all those with party sympathies vote, and that the propensity to vote varies from one party to another. Sifo's election poll takes no account of voting propensity in stating figures for the various parties. The propensity to vote among people with party sympathies is tabulated separately, and published only during four quarters before an election.

Deviation from trend (margin of error)

The margin of error is the same as the standard deviation of results obtained from measurements from numerous repetitions with new random samples. The standard deviation of the difference between the election poll figures and the trend figures is our equivalent of mean error. After 112 election polls, we have drawn the conclusion that one out of 20 deviations from trend exceeds the following figures for the parties concerned:

1.11% for the Moderate Party
1.51% for the Liberal Party
1.60% for the Center Party
1.53% for the Social Democratic Party 0.90% for the Left Party Communists
(See page 21)

Sifo has rounded these figures according to a rule of thumb for mass media and the public that is published with every election poll: "Attention should not focus on changes from a previously established trend or level that is less than 2 percent for parties with over 20 percent of the party sympathizers, or less than 1.5 percent for the other parties." We consider it correct that the rule of thumb is somewhat stricter than the above figures would suggest, since the published figures are rounded to the nearest whole or half of a percentage point. A mistake that occurs frequently in the mass media is to calculate deviations from the previous measurement instead of from a previously established trend or level.

The trend and deviation from trend are shown graphically on pages 9-18. Calculations of their standard deviations are given on pages 19-21.

Deviation from sampling plan (nonresponse)

The weekly omnibus uses multi-stage sampling (described separately), which is random at all stages, designed according to the principles of optimal stratification. The omnibus survey (and election poll) are also poststratified.

Nonresponse, respondent dropout from the survey population, and no answers are reported on pages 22-24.

New interviewees are selected for each survey occasion. The samples are additive.

In 1978, the basic master sample comprised 380 postal codes. The names and addresses of the interviewees selected are chosen at random from the population register. The deceased; persons who have emigrated or spent spend a long period abroad or at sea, are inmates of institutions (prisons, mental hospitals, etc); persons who are ), senile, deaf-mute, or severely ill; persons who have moved outside the interview catchment area or to an unknown address; soldiers on maneuver; and also persons who are unable to speak Swedish reasonably well are excluded from the survey population. It is not correct or possible for Sifo's staff to interview these people. Many of those who are excluded from the survey population are entitled to vote. The survey population does not include all those who are eligible to vote, but only those voters who are, basically, to be interviewed. The proportions of the total number of names excluded from the survey population have, in recent years, been as follows:

1972       14.2%
1973       14.8%
1974       15.9%
1975       14.3%
1976       12.7%
1977       10.6%
1978       10.5%

The downturn from 1976 is connected with the fact that the upper age limit for the sample was reduced to 70 years (See the note on page 3).

Nonresponse includes persons who refuse to be interviewed or whom the interviewers were unable to contact; those who had no time for an interview on the dates the interviewer was able to offer; those who broke the interview agreements; and those who were not interviewed for other reasons. Thus, nonresponse constitutes all the people whom Sifo's staff, in principle, should have been able to interview but were not interviewed. Nonresponse in every election poll since 1972 is reported on pages 22-24 and graphically on page 25. [Från 1967 till juni 1970 substituerade Sifo borfallet. Se not på sid 24]

Summing up, we have had the following nonresponse rates since 1972:

1972       18.7%
1973       18.8%
1974       19.4%
1975       20.2%
1976       21.0%
1977       21.4%
1978       20.8%

Eight-tenths of the nonresponse are people who refuse to be interviewed. We must respect their integrity.

On pages 22-24  the rate of “no answer" is also reported, i.e. interviewees who were unable to answer questions about their preferred party, or did not have a "best party" at all. There was a decisive downturn in no answer in spring 1973, when secret ballot-papers became the rule in Sifo's election polls.  [See the note on page 24]. Information on this matter and poststratification in the election poll may be found in the report  "Using Data on Party Support in the Last Election in Polls of Party Sympathies between Elections. Some SIFO Findings" by Hans L Zetterberg and Karin Busch.

Pages 26-35 contain regression analyses and variance analyses of the effects of nonresponse on the parties' positions in the election poll. One consistent conclusion is that such effects cannot be proved. In the equation

Party share = P*(nonresponse) + Q*(date) +  constant

P is always close to zero and never anywhere near statistical significance. The same applies, moreover, if we calculate changes in party shares and deviation from trend.

No party appears to be a systematic winner or loser from higher or lower nonresponse within the range in which election-poll nonresponse varies. Accordingly, we do not wish to rule out the possibility of particular political events making a specific party's electors more or less inclined to take part in an opinion poll.

January 18th, 1979

TABLES

Sifo's election polls, 1967--78. Party ranking expressed as percentages of interviewees with party sympathies   1
Note on upper age limit in the election poll   3

Party ranking in Sifo's election polls, 1967-78, expressed as trend values. Henderson's weighting of a nine-month trend cycle 4

Computer graph of party ranking, 1967-78, expressed as trend values   7

Computer graph of percentage support for The Right Party/Moderate Party and its deviation from trend, 1967-78   9

Computer graph of percentage support for Liberal Party and its deviation from trend, 1967-78   11

Computer graph of percentage support for Center Party and its deviation from trend, 1967-78   13

Computer graph of percentage support for Social Democratic Party and its deviation from trend, 1967-78   15

Computer graph of percentage support for Left Party Communists, 1967-78 and its deviation from trend   17

Sifo's election poll: deviations from party trends, 1967-78, expressed as percentages   19
Summary   21

No answers ("don't know"), nonresponse (no interview) and number of interviewees in Sifo's election poll, 1967-78   22
Note on use of substitutes in the event of nonresponse, 1967-72   24

Computer graph of nonresponse trend, 1972-78   25

Regression of the effects of nonresponse on Moderate Party's share in the election poll, 1972-76   26-27

Regression of the effects of nonresponse on Liberal Party's share in the election poll, 1972-76   28-29

Regression of the effects of nonresponse on Center Party's share in the election poll, 1972-76   30-31

Regression of the effects of nonresponse on Social Democratic Party's share in the election poll, 1972-76   32-33

Regression of the effects of nonresponse on Left Party Communists' share in the election poll, 1972-76   34-35

 

QUESTIONNAIRE

Cover of questionnaire in Sifo's weekly omnibus   36-38

Questionnaire and ballot-papers for Sifo's election poll   39