Poll oversamples minority students

Despite oversampling the population of minority students attending the university, Blue Hen Poll organizers say the questionnaire’s results more accurately represent the opinions of the school’s undergraduate community than in previous years.

In order to gain a more accurate understanding of the student body, political science professor David Wilson, faculty advisor for the research team that conducted the poll, said the questionnaire was sent to a random sample of 2,741 white and international students, and all of the 2,259 minority undergraduate students attending classes at the university campus. Previous years’ polls were only sent to a sample of 2,500 random students.

Wilson said the sample of minority students polled had to be larger than a random sample of all students at the university due to the fact that the entire student body is 75 percent white. He said if researchers were to choose a student at random to interview there would only be a 30 percent chance that student is not white.

In previous versions of the poll, he said approximately 60 to 70 out of 1,000 respondents were minority students. The low percentage indicates results from those polls may not gauge the opinions of minority students as accurately as this year’s poll, he said.

Wilson said this was the first time the Blue Hen Poll oversampled racial minorities. However, three years ago the poll used statistical weighting for graduate students’ responses, assigning a number to each value of a given quantity and giving the number of times this value is found to be observed.

Wilson said weighted responses were also used in this year’s poll to accurately represent the student body.

“You need to statistically adjust those numbers to make it look like the real world,” he said. “We call that statistical weighting.”

Wilson said they also wanted to make a significant change to the questionnaire for the fifth anniversary of the Blue Hen Poll.

“This year it was important to be more representative of what the university really looks like,” he said.

Larger sample sizes of small groups provide smaller margins of error, Wilson said.

“The goal for us was to be able to make comparisons across race and ethnicity, and to do that we have to get large sample sizes for each race and ethnicity,” he said.

Food and resource economics professor Thomas Ilvento, who teaches a class about research methods, said oversampling and weighting data are tools that can produce more accurate results.

Ilvento said oversampling small populations within the university community when compiling a survey is useful in making comparisons to larger populations.

Of the 1,607 respondents, 493 were racial minority and 728 were white. The numbers were weighted to demonstrate actual proportions which were 78 percent white, 4 percent black, 6 percent Hispanic, 4 percent Asian and 8 percent other. The male to female ratio of respondents was 43 males to 57 females.

Illvento said making a statement regarding the overall student population while oversampling African-Americans, without adding weight to their responses, would hypothetically decrease a survey’s accuracy.

“The excess number of African-Americans will bias or distort the result,” he said. “So now I use a weighting scheme to make the oversampled group’s responses have less weight in the overall sample.”

Graduate student Andrew Hellwege, a teaching assistant for the research team, said organizers worked with the Office of Institutional Research and Planning to compile the sample of 5,000 students who received the poll.

Hellwege said the research team used an online survey to distribute the poll because it was inexpensive to administer and convenient for researchers and respondents. He said the research team achieved their target number of respondents using this method.

Hellwege said the response rate of the 5,000 students in the sample was 32 percent, which is about double the average response rate of other surveys administrated by the university, such as Student Health and Dining Services. Research team members used approximately 1,312 of the responses to analyze data, which gave them a margin of error of three percent when analyzing data, he said.

“We are 95 percent confident that the results of our survey fall within plus or minus three percentage points of the true population,” Hellwege said.

The margin of error describes the whole study, rather than individual questions, Wilson said. Some questions, such as one which asked students about the positive or negative interactions with Newark or campus police, had approximately 6 percent of respondents skip the question, which is much higher than others, he said. The first question about overall satisfaction with the university as a whole was skipped by 1.1 percent of respondents.

Wilson said research team members expect approximately 5 percent of omitted answers are the result of technical issues, such as losing Internet connection, not completing the survey or choosing to not respond. He said skipped questions do not interfere with the way the research team analyzes data.