本研究考慮到影響選民投票行為的三大因素：政黨、政見/議題、候選人，並運用三個預測變數：「候選人形象評價」、「選民預期候選人解決問題的能力」以及「選民的政黨支持」，來作為預測1996年總統選舉各組候選人的得票率，預測結果與實際開票結果極為接近，各組候選人得票率的預測差距皆在三個百分比以內。而且自三波調查資料來看，選前十天所作的選舉預測與選前一星期及選前三天所做的預測結果，並沒有太大的差距，顯示此預測模型有其一定的穩定性。 就本研究而言，如果本預測模型能夠禁得起考驗的話，其對台灣地區總統選舉選民投票行為的貢獻是在於民意調查中未表態選民之投票行為規則的提出：一、如果未表態選民能夠在候選人的形象評價上評出最高分，則其會投給相對應的候選人；二、如果未表態選民不能在候選人形象評價上評出最高分，則其會視候選人未來解決問題的能力高低而定；三、如果未表態選民能夠對於候選人未來解決問題的能力分出高下，則其會投給相對應的候選人，否則會考慮候選人所代表的政黨；基本上會傾向投給候選人所署政黨與未表態選民本身認同政黨一致的候選人 ( 但此情況並非那麼確定，只能說是一種傾向 ) ；四、經前述一至三程序仍未能決定投票對象的未表態選民，基本上會比較傾向支持民進黨的候選人，但此規則之經驗基礎非常薄弱，仍待進一步的探討研究。

By using the following three variables: candidate image evaluation, expected capability in solving problem, and voter's party preference in predicting in 1996 presidential, this study demonstrates that the findings have been highly close to election result, with the difference falling within three percent. The model has also been proved stable in terms of its minor variable in prediction throughout the three different testing time frame. The major findings of this study can be summed up as following: first of all, a no-response voter could vote for the candidate who was ranked first in image evaluation; secondary, problem-solving capability will be the second electoral decisive factor for a no-response voter, if he/she could not decide which candidate scores highest; thirdly, if a no-response repondent can decide which candidate could be the most capable one, he/she would vote accordingly, otherwise he/she tends to vote for the candidate with the same party affiliation; fourthly, those could not be screened out through the preceding process are, to an extent, inclined to vote for the DPP's candidates than for the counterparts of the rest two parties.