Presentation #1:

"Non-Response Bias: Recent Findings from our Address-based Panel Recruitment"
by Joe Garrett and Mike Dennis

The presentation focuses on statistical tests of sample representativeness of ABS-sourced sample for KnowledgePanel members who are randomly selected from the USPS Delivery Sequence File.

Address-based sample (ABS) frames have emerged as a solution to shrinking RDD frame coverage due to cell phone-only households.

The Non-Response Analysis tested:

• Extent of self-selection bias by comparing the recruited cooperating sample to the full invited ABS sample on demographic, socio-economic, and other variables hypothesized to be correlated with survey estimates.
• Recruited sample demographics as compared to the U.S. Census and NHIS benchmarks.
• Ecological analysis was prepared to assess the propensity of recruited cases to have voted for Barack Obama in the general election, compared to the propensity of non-responders.

Many of the analyses we conducted are enabled by the fact that ABS recruitment data can be linked to the home address for imputing household characteristics.

The analysis confirmed:

• ABS-sourced panel recruits are reasonably comparable to the NHIS benchmarks for proportion of cell phone-only households, gender, and age.
• In comparing recruited sample to the full-invited sample, ABS recruits have a remarkable comparability on secondary demographics.
• Propensity to join KnowledgePanel was almost identical for the ABS recruits compared to the non-respondents in regard to voting behavior in the 2008 Presidential election.

Future research will be conducted using additional supplemental variables.

Presentation #2:

"Does Providing Internet Access to Non-Internet Households Affect Reported Media Behavior for Latinos and Non-Latinos?: Results from a Six-Month Longitudinal Survey"
by Mike Dennis and Pat Graham

Previous research conducted by Knowledge Networks has shown few significant effects on Web panel survey results resulting from "panel conditioning," that is, changes in survey responses due to prior experience in participating in panel surveys.

The Current Conditioning Analysis tested:

• Whether, and in what ways, panel conditioning is operative in the KnowledgePanel Latino (KPL) sample, since KN provided a laptop and free ISP to roughly 55% of these new members.
• Whether the provision of a laptop and Internet service is having an impact on survey responses to media behavior questions.
  • We hypothesized that we would observe a short-lived "novelty effect" for the new panelists given Web devices – an early "excitement" about receiving the Web device – followed by a return to previous levels of media behavior.

Design:

• Three-wave longitudinal study conducted with newly recruited KnowledgePanel Latino panelists.
• Feb '09 baseline survey with the same survey questions administered to the same respondents in two successive waves of data collection (at 3-months and 6-months post baseline).
• Analysis focuses on 150 Latino cases who participated in all three waves of data collection, and to a lesser extent, on the 62 non-Latinos participating in the experiment.

Analysis Results:

• Supporting our initial hypotheses, we observe no significant behavior changes among non-Internet households who were given laptops in terms of traditional media usage and online behavior.
• In the few instances where we measured some change, the direction of the change is showing lower media usage at waves 2 and 3 in comparison to the baseline survey, consistent with the "novelty effect" hypothesis that predicts lower levels of media usage after the initial excitement in joining the panel.
• The "novelty effect" was most substantively observed for frequency of Web site visits.
• The large majority of measures showed no change over the three waves of data collection – consistent with the hypothesis that panel conditioning is not operative as a general mechanism.