《纽约时报》/锡耶纳学院对阿拉斯加、爱荷华、俄亥俄、德克萨斯和堪萨斯的民调是通过手机和固定电话进行的。
The New York Times/Siena polls of Alaska, Iowa, Ohio, Texas and Kansas were conducted on cellular and landline telephones.
爱荷华州606名选民和俄亥俄州616名选民的民调于2026年9月22日至10月1日以英语进行。德克萨斯州615名选民的民调于9月21日至9月30日以英语和西班牙语进行。堪萨斯州605名选民的民调于9月22日至9月30日以英语进行。阿拉斯加州504名选民的民调于9月24日至10月1日以英语进行，并包括少量收到短信后在网上参与调查的受访者。
The polls of 606 voters in Iowa and 616 voters in Ohio were conducted in English from Sept. 22 to Oct. 1 2026. The poll of 615 voters in Texas was conducted in English and Spanish from Sept. 21 to Sept. 30. The poll of 605 voters in Kansas was conducted in English from Sept. 22 to Sept. 30. And the poll of 504 voters in Alaska was conducted in English from Sept. 24 to Oct. 1 and included a small number of respondents who took the survey on the web after receiving a text message.
在可能的选民群体中，抽样误差范围为：爱荷华州和堪萨斯州为正负4.6个百分点；阿拉斯加州为正负5.1个百分点；俄亥俄州和德克萨斯州为正负4.5个百分点。合并分析各州时，抽样误差范围为正负2.1个百分点。
The margin of sampling error among the likely electorate is plus or minus 4.6 percentage points in Iowa and Kansas; plus or minus 5.1 percentage points in Alaska; and plus or minus 4.5 percentage points in Ohio and Texas. The margin of sampling error when analyzing the states together is plus or minus 2.1 percentage points.
《纽约时报》/锡耶纳民调是《纽约时报》与锡耶纳研究院（ReconMR旗下机构）的合作项目。
The Times/Siena Poll is a collaboration between The New York Times and Siena Research Institute, part of ReconMR.
抽样方法：本调查采用响应率调整、分层、概率比例抽样（PPS）方法，从L2维护的选民档案（非党派选民档案供应商）中抽取可能参加2026年选举的选民样本，并补充了营销系统集团提供的经选民档案匹配的额外手机号码。《纽约时报》通过多步骤选择样本，以考虑不同的电话覆盖率、非响应情况以及各州电话号码生产力的显著差异。
Sample The survey is a response rate-adjusted, stratified, probability-proportional-to-size sample of the likely 2026 electorate taken from the voter file maintained by L2, a nonpartisan voter-file vendor, and supplemented with additional voter-file-matched cellular telephone numbers from Marketing Systems Group. The sample was selected by The New York Times in multiple steps to account for differential telephone coverage, nonresponse and significant variation in the productivity of telephone numbers by state.
为调整非覆盖偏差，选民档案按州众议院选区、党派、种族、性别、婚姻状况、家庭规模、投票历史、年龄和住房所有权进行了分层。计算了每个层中拥有电话号码的登记选民比例及平均预期回复率。平均预期回复率基于过往《纽约时报》/锡耶纳调查的单位无应答模型得出。初始选样权重等于某层平均电话覆盖率与建模回复率乘积的倒数。对于L2档案中拥有多个电话号码的受访者，或L2与Marketing Systems Group档案号码不一致的受访者，选择建模回复率最高的号码。
To adjust for noncoverage bias, the voter file was stratified by statehouse district, party, race, gender, marital status, household size, turnout history, age and homeownership. The proportion of registrants with a telephone number and the mean expected response rate were calculated for each stratum. The mean expected response rate was based on a model of unit nonresponse in prior Times/Siena surveys. The initial selection weight was equal to the reciprocal of a stratum’s mean telephone coverage and modeled response rate. For respondents with multiple telephone numbers on the L2 file, or with differing numbers from L2 and Marketing Systems Group, the number with the highest modeled response rate was selected.
实地调查 样本按政党、种族和地区进行分层。Marketing Systems Group对样本进行筛选，确保手机号码处于活跃状态。锡耶纳大学锡耶纳研究所和Reconnaissance Market Research (ReconMR) 实施了民调。访员要求与选民档案上姓名一致的人通话，若预定受访者不在则结束访问。总体而言，各州超过98%的受访者通过手机接触到。
Fielding The sample was stratified according to political party, race and region. Marketing Systems Group screened the sample to ensure that the cellular telephone numbers were active. The Siena Research Institute at Siena University and Reconnaissance Market Research (ReconMR) fielded the poll. Interviewers asked for the person named on the voter file and ended the interview if the intended respondent was not available. Overall, more than 98 percent of respondents in each state were reached on cellular telephones.
在阿拉斯加，初步电话尝试后，建模种族为“其他”的剩余受访者样本收到包含在线调查参与邀请的短信，获得一份回复。
In Alaska, after initial attempts by phone, the remaining sample for respondents with an “other” modeled race were sent a text message containing an invitation to participate in an online survey, yielding one response.
在阿拉斯加，因9月24日发现实地调查错误，民调予以重置。有错误的回复被丢弃，已尝试联系的选民也被剔除，配额重新设定。
In Alaska, the poll was reset after a fielding error was found on Sept. 24. Responses with the error were discarded, as were voters who had already been attempted, and quotas were reset.
如果受访者在被询问用于加权的三个自报变量——年龄、教育程度和回忆的2024年投票情况——后没有退出调查，并且回答了关于年龄、教育程度或2026年大选投票意向的至少一个问题，则该访谈被视为完整，可纳入关于受访者将投票给谁的问题中。
An interview was determined to be complete for the purposes of inclusion in the questions about whom the respondent would vote for if the respondent did not drop out of the survey after being asked the three self-reported variables used in weighting — age, education and recalled 2024 vote — and answered at least one of the questions about age, education or 2026 general election vote intent.
加权 《纽约时报》使用R语言中的WeightIt包，对调查进行了多步骤加权。首先，根据层别调整了样本的不等选择概率。
Weighting The survey was weighted in multiple steps by The Times using the WeightIt package in R. First, the sample was adjusted for unequal probability of selection by stratum.
其次，使用能量平衡法对样本进行加权，该方法通过最小化调查样本与目标总体之间的能量距离——一种衡量两个多元分布差异的统计量——来确定权重。
Second, the sample was weighted using energy balancing, which finds weights by minimizing the energy distance — a statistical measure of the difference between two multivariate distributions — between the survey and the target population.
目标总体是从L2选民档案的活跃选民名单中抽取的可能选民分层样本（n=40,000）。活跃选民在2026年中期选举中投票的概率，基于2022年和2018年中期选举经验证的投票率模型。调查通过加权来平衡以下特征的联合分布：党派（如该州提供党派登记信息则采用该信息，否则依据该州提供的党派初选参与情况分类，否则依据过往《纽约时报》/锡耶纳民调中的投票选择模型分类） 种族（L2种族模型） 性别（L2数据） 受教育程度（自我报告的四个教育水平类别，加权以匹配基于《纽约时报》/锡耶纳民调、人口普查数据及L2选民档案得出的《纽约时报》目标值） 投票历史（《纽约时报》基于L2数据的分类） 州区域（《纽约时报》按县或城市划分的分类） 合成2024年过往投票 2024年总统选举投票选择的模型概率（《纽约时报》基于过往《纽约时报》/锡耶纳民调中的投票选择模型，依据L2选民档案上的可用信息并按选区级选举结果调整得出） 第三，权重经过调整以纳入自我报告的投票意向，该调整基于过往《纽约时报》/锡耶纳民调受访者中经验证的投票率模型，作为受访者自我报告投票率及其选前投票率模型的函数，并包含一个惩罚项，以解释调查受访者比非受访者以更高比例投票的倾向。
The target population was a stratified sample (n=40,000) of the likely electorate, drawn from the L2 voter file’s list of activeed voters. The probability that an activeed voter will turn out in the 2026 midterm election is based on a model of validated turnout in the 2022 and 2018 midterm elections. The survey was weighted to balance the joint distribution of the following characteristics: Party (Party registration if available in the state, else classification based on participation in partisan primaries if available in the state, else classification based on a model of vote choice in prior Times/Siena polls) Race (L2 model of race) Gender (L2 data) Education (four categories of self-reported education level, weighted to match NYT-based targets derived from Times/Siena polls, census data and the L2 voter file) Turnout history (NYT classifications based on L2 data) State regions (NYT classifications by county or city) Synthetic 2024 past vote Modeled probability of presidential vote choice in the 2024 election (NYT model of vote choice in past Times/Siena polls based on information available on the L2 voter file and adjusted to precinct-level election results) Third, the weights were adjusted to incorporate self-reported intention to vote, based on a model of validated turnout among prior Times/Siena poll respondents as a function of a respondent’s self-reported turnout and their pre-election turnout model, including a penalty to account for the tendency of survey respondents to turn out at higher rates than nonrespondents.
投票可能性高于其初始模型投票概率的受访者获得更高权重，最终权重等于初始权重乘以最终投票概率再除以先验模型投票概率。
Respondents with a higher likelihood to vote than their initial modeled turnout probability receive more weight, with the final weight equal to the initial weight, multiplied by the final turnout probability and divided by the ex ante modeled turnout probability.
最后，完成所有调查问题的受访者样本采用能量平衡法进行加权，以匹配上述完整调查样本特征的联合分布，以及美国参议院选举的结果。
Finally, the sample of respondents who completed all questions in the survey was weighted using energy balancing to match the joint distribution of the aforementioned characteristics of the full sample of the survey, as well as the results of the race for U.S. Senate.
误差范围 误差范围考虑了调查的设计效应，即衡量因调查设计和加权导致的统计效力损失的指标。
Margin of error The margin of error accounts for the survey’s design effect, a measure of the loss of statistical power due to survey design and weighting.
完整样本的设计效应为：爱荷华州可能选民群体为1.35，阿拉斯加州为1.39，俄亥俄州为1.30，德克萨斯州为1.28，堪萨斯州为1.32，五州合计为1.33。
The design effect for the full sample is 1.35 for the likely electorate in Iowa, 1.39 in Alaska, 1.30 in Ohio, 1.28 in Texas, 1.32 in Kansas, and 1.33 across all five states.
对于完成访谈的样本，在可能选民群体中，抽样误差范围为：爱荷华州正负5.2个百分点，含设计效应1.34；阿拉斯加州正负5.5个百分点，含设计效应1.38；俄亥俄州正负5个百分点，含设计效应1.30；德克萨斯州正负5.2个百分点，含设计效应1.33；堪萨斯州正负4.9个百分点，含设计效应1.31；五州合计正负2.3个百分点，含设计效应1.33。
For the sample of completed interviews, among the likely electorate, the margin of sampling error is plus or minus 5.2 points in Iowa, including a design effect of 1.34; 5.5 points in Alaska, including a design effect of 1.38; 5 points in Ohio, including a design effect of 1.30; 5.2 points in Texas, including a design effect of 1.33; 4.9 points in Kansas, including a design effect of 1.31; and 2.3 points across all five states, including a design effect of 1.33.
从2016年到2024年，《纽约时报》/锡耶纳学院民调在选举前最后三周进行的调查中，第95百分位的误差为正负5.6个百分点。现实世界中的误差包含除抽样误差以外的误差来源，如无应答偏差、覆盖误差、犹豫不决选民的晚期转向以及对选民构成估算的误差。
From 2016 to 2024, The Times/Siena Poll’s error at the 95th percentile was plus or minus 5.6 percentage points in surveys taken over the final three weeks before an election. Real-world error includes sources of error beyond sampling error, such as nonresponse bias, coverage error, late shifts among undecided voters and error in estimating the composition of the electorate.