The spurt lasts three weeks but it wins for you - when you might have lost by a heart-breakingly small margin. Perhaps you win by a fat margin and perhaps the spurt was not really necessary - you will never know but it does not matter; you've won.
Suppose your poll shows a conservative estimate of more than 50%; you are then justified in continuing your present campaign plans, without an emergency spurt but without slackening off.
Suppose the poll had been the other way around, 22 for Upright; 28 for Hopeful - your conservative estimate is then 36%. Does this mean you should quit? No, for Hopeful's conservative estimate is still less than 50%. It means a tough fight with a possibility, but not much probability, of winning. Stick with it.
Suppose Hopeful got 30 votes in the poll, indicating that he will probably beat your man by at least 52% of the vote and that he might take as high as 68% of the total. Should you throw in the sponge?
Not on your own initiative - I recommend that you talk it over with your candidate, then call a closed meeting of all workers and all money contributors, tell them the sad news and ask them to express their wishes. From a cold-blooded standpoint you might as well cut your losses and quit... but I predict that they will vote to stick to the finish and turn the meeting into a rally. They may even win for you. Politics isn't dice, nor statistical physics; the Spirit of the Alamo may outweigh all measurable factors.
If they decide to stick, bow to their will and pitch in. It will be a treasured emotional experience at least - and many a "lost" campaign has planted the seed for an eventual political upheaval.
The Meaning of "Random": A "random" sample is one which is as truly representative of the district as you can make it. This is most easily done by trying to keep out the personal element in the selections. For example - you want 100 names from 200 precincts: Take the bottom name, of your party, from the second column of each-even numbered precinct list. Or make up any other rule which makes the selection mechanical, with no choice on the part of the operator, and which spreads the sample evenly through the district, according to population, not area.
Never take the sample all from one precinct or one area. If you are polling by telephone you will find that some of your choices do not have telephones. Do not substitute the next name having a telephone listing; the voters without telephones must be polled at their homes - otherwise you will introduce an economic factor which will falsify your answer.
Polling by telephone is best done in the evenings, in order to find both men and women at home. Do not accept the response of a spouse in place of the voter named by the random choice; it will change your results... there is a definite tendency for women to vote more conservatively, and in other ways differently, than do men.
Do not let the polling question suggest the answer desired. For example, here is a suitable phraseology for a telephone polclass="underline" "Good evening, is this Mrs. Mabel Smith? Mrs. Smith, this is the civic affairs research bureau speaking. Have you formed an opinion about the congressional candidates who will appear on your primary ballot a week from next Tuesday?"
It should be possible for one worker to prepare a list for a telephone poll in one evening and get fifty responsive answers in not more than three evenings. A reply-postal card poll should take about the same length of time to prepare and is about as accurate, but it takes longer to get the results and 250 should be the minimum sent out. It may be cheaper than telephoning in districts involving long-distance tolls. (These reply-type postal cards, at two cents apiece, are invaluable in penny-pinching political work.)
Don't attempt to make a straw-vote canvas door-to-door. Don't try it on the street. The names mud be pre-selected by some non-personal method. Mathematical Basis f or the Rule-of-Eight: (Skip this, if you like.) In any statistical sampling the larger the sample, the smaller the errors in the result, except for systematic errors - errors which are inherent in the thing being sampled. In the opinion of this writer, the systematic errors in any poll of political opinion conducted without expert actuarial help are so large that it is not worth while to use a sample larger than 100. On the other hand the "probable errors" - errors which depend on the laws of chance - are so large for samples less than 50 that trends will be masked by the inescapable "probable errors." For efficient use of time and money the smallest sample which will spot a trend is desired. For that reason, and because percentages may be obtained from a 50-sample simply by doubling (percentage problems are troublesome to some), a sample of 50 has been recommended.
Bessel's formula for probable error has been used in computing the rule-of-eight, assuming independent events of equal probability and assuming a "universe" of very large but limited numbers. The assumption of equal probability may be attacked; the pragmatic justification lies in the fact that probable errors are largest in a 50-50 division and the political situation is most critical in such a situation - a landslide either way will show in a sample of 50 without resort to probable error. The rule-of-eight is neither the "probable error"of the engineer, nor the three-standard-deviations-equals-standard-certainty of the professional statistician; the first was rejected as too esoteric in meaning for the layman, the second was rejected because trend-spotting with it requires samples too large for the volunteer political campaign. A selected error of 8% was chosen to produce a conservative probability of about four-to-one, which was considered accurate enough for the purpose and much more reliable than most data we plan our lives by - in choosing a wife, for example!
If greater accuracy can be afforded, use a sample of 100 and a rule-of-five. Or the mathematical reader may perform his own analysis, following Peters or Bessel or others; I can't recommend direct analysis using the binomial expansion without pre-computa-tion, even using Pascal's triangle - the figures are incredibly astronomical!
Sampling by "Smell": In addition to poll-taking and making predictions, try this-in time you will acquire skill in it: Prowl through your district Buy a Coke and chat with your druggist. Buy two gallons of gas-chin with the man at the pumps. Ask strangers for matches, then gossip. Get a haircut. Make a purchase in an uncrowded grocery. Ask passing strangers for information-then talk.
When you have done this you will combine it subconsciously with the doorbell punching you have done (which, for the manager, should be scattered through the district) and you will end up with a curious feeling way down inside. Drag it up and into the light, take a look at it, and see whether or not it tells you that your man is going to win.
The human mind, when trained, is capable of more rapid, more flexible, and more reliable evaluations of problems containing unlimited unknowns than any of the mechanisms as yet invented. In time you will acquire this talent; you will know it when your predictions are consistently correct, not only as to results but as to approximate majorities and size of vote cast.
The acquisition of the talent is painless and almost effortless.
While acquiring the talent don't let yourself be panicked by some phony figures. Amateurs are inclined to think that their strenuous efforts must be producing a tidal wave, then are disappointed when they go out on a "sniffing" tour and find hardly a ripple. That is normal; primary campaigns hardly ever stir the general public out of their sleep. All you need is a ripple, of the right size, and in the right place. You know it is in the right place for you have been using the direct vote-getting methods; now you want to know if it is the right size.