Based on the demographic covariates included in , the study population was similar in all 3 years and reflected the Medicare population.
Association between Media Reports and Vaccination Timing
shows the multivariate associations between vaccination timing and the mass media reports. Columns 1, 2, and 4 in show the associations when a media report contained keywords of “influenza” or “flu.” Other columns included additional keywords of “vaccine” or “shot” and of “delay” or “shortage” or “late.”
Table 2 Effects of Mass Media Reports of “Influenza” and “Influenza Vaccine Delay/Shortage” on Influenza Vaccination Timing among Medicare Elderly (Years 1999, 2000, 2001) Estimated by Generalized Gamma Model (Coefficient [SE]) (more ...)
Each model included three types of media variables measuring the common set of keywords appearing in the headline, except a model containing keywords of “delay/shortage” in 2001 where keywords appeared in the text only. Because the headline tended to have a stronger association with vaccination than keyword matches in the text only, models in included the media variables of headlines unless headlines were unavailable. These models used the periods of measuring media reports with a stronger association with vaccination, unique to each year. Namely, the models in 2001 used the media reports measured during an immediate past 1 week (1–7 days), and the models in 1999 and 2000 did the media reports during a past 1 week with a 1-week lag (8–14 days).
It should be noted that the coefficients of gamma models in indicate the proportion of “time spent in the unvaccinated state over a period of 4 months conditional on being exposed to an influenza-related report with a circulation/audience households (AHH) of one million” divided by “the same time conditional on no report exposure.” That is, these coefficients are normalized to measure the effect of changing circulation/AHH by 1 million. The actual circulation/AHH numbers differ from that. Also, a covariate's coefficient being smaller than one implies the positive association between the covariate and earlier vaccination timing.
For example, the coefficient of 0.981 for USA Today in 2001, in the second row in the far right column in , indicates that unvaccinated time of a Medicare beneficiary with the exposure of one USA Today article containing “influenza vaccine delay/shortage” in the past week is estimated to be 98.1 percent of unvaccinated time without such exposure. Because the actual circulation of USA Today in 2001 was 2.36 million, the actual effect on vaccination timing or the proportion of unvaccinated time was not 0.981 (per 1 million circulation) but 0.956 (=0.9812.36), which is statistically significant (p<.01). Similarly, because on the average 4.25 million households tuned to CBS network in 2001, one CBS network program reporting “influenza vaccine delay/shortage” reduced the proportion of unvaccinated time to 0.971 (=0.9934.25), based on the 4TV coefficient of 0.993 (third row, far right column in ).
The association between receipt of influenza vaccine and number of any type of media reports was multiplicative, so that reports from multiple sources during the same week were associated with greater likelihood of vaccination. Extending two examples described in the previous paragraph, when “influenza vaccine delay/shortage” was reported by one USA Today article and one CBS network program during the same week in 2001, the proportion of unvaccinated time was 0.928 (=0.956 (USA Today) × 0.971 (CBS) shown in the previous paragraph), that is, the vaccination timing became much earlier than a case exposed to single media source.
Among all models in , all three media variables were positively associated with earlier vaccination timing. These associations were statistically significant (p<.05) for 4TV in all five models and for USA Today and AP in three out of five models.
Effects of Additional Keywords
Adding the keywords “vaccine” or “shot” and “delay” or “shortage” or “late” tended to increase the association between a media report and vaccination timing in terms of magnitude or statistical significance level, regardless of the type of media. Coefficients in columns 3 and 5 are smaller than or equal to corresponding coefficients in columns 2 and 4 in (p<.05).
Effects of Controlling Variables
CDC press releases had inconsistent associations with the vaccination timing. The association between vaccine availability and vaccination timing was always positive and statistically significant (p<.01) in all models. For example, the model in 1999 in (row 6, column 1) implied that an additional 1 million vaccine doses distributed in a resident region over the past 4 weeks was associated with an unvaccinated period reduced by 5.2 percent (=1−0.948), relative to a region in which no additional vaccines were distributed.
As discussed earlier, we present two hypothetical extreme settings where media affected either vaccination timing only () or annual-rate only (). In reality, media is expected to have both effects but in a smaller magnitude than those presented in these tables.
Table 3 Effects of Mass Media Reports of “Influenza” and “Influenza Vaccine Delay/Shortage” on Mean Influenza Vaccination Timing† among the Vaccinated Medicare Elderly (Year 1999, 2000, 2001) Estimated by Generalized Gamma (more ...)
Effects of Mass Media Influenza-Related Reports on Annual Influenza Vaccine Receipt† among Medicare Elderly (Year 1999, 2000, 2001) (% change in annual vaccination rate) (based on generalized-gamma model in )
Association between Media Reports and Vaccination Timing Measured by Days Received Earlier
presents the shift of mean vaccination timing among the vaccinated associated with media reports, corresponding to , which presented gamma model coefficients. presents the media effects in a hypothetical scenario (columns 1, 3, and 5) and those in an observed scenario (columns 2, 4, and 6). In one of the hypothetical scenarios (row 1, column 5 in 2001), if USA Today reports one headline article on “influenza” every week from September 1 to December 31, the mean vaccination timing is expected to shift earlier by 15.3 days (15.3=(3.18)2.36, assuming 2.36 million circulation) compared with a hypothetical season without such headline at all. In fact, such headline articles (seven in total) appeared in 5 weeks in 2001 and were associated with the shift in the mean vaccination timing by 2.05 days earlier (row 1, far right column in 2001).
Comparing the magnitude of 1 unit increase of a certain media variable across 3 years, the magnitude in 2000 tended to be greater. However, the magnitude of observed reports tended to be smaller in 2000 than in 2001, probably due to the most serious vaccine supply limitation in 2000. It should be noted that the mean vaccination timing would be delayed, if, for example, media reports in late December increased the annual vaccination rate but also increased the proportion of late vaccination timing among the all vaccinated.
Association between Media Reports and Annual Vaccination
presents the multivariate associations between annual vaccine receipt and mass media reports, corresponding to . Influenza-related mass media reports were positively associated with the annual vaccination rate. Like , also presents the media effects in a hypothetical scenario and those in an observed scenario. In one of the hypothetical scenarios (row 1, column 5 in 2001), if USA Today reports one headline article on “influenza” every week from September 1 to December 31, the annual vaccination rate is expected to increase by 9.3 percentage points (1.093=(1.0383)2.36, assuming 2.36 million circulation) compared with a hypothetical season without such headline at all. Actual headline articles (seven in total) in 2001 were associated with the increase in the annual vaccination by 5 percentage points (row 1, far right column in 2001), in comparison with a scenario where those headlines did not occur.
As observed in , the positive association between annual vaccination and the media reports tended to be greater, particularly in 4TV and USA Today, when a report included additional keywords like shortage/delay. Comparing the magnitude of a 1 unit increase of a certain media variable across 3 years, the magnitude in 2000 tended to be greater. However, the magnitude of observed reports tended to be smaller in 2000 among 3 years, probably due to the most serious vaccine supply limitation in 2000.
In secondary analysis, logit models yielded unstable and implausibly large estimates compared with those of gamma models.