Sampling Overview

Excerpt from the Book:

"Table 3.1 shows the total number of articles we found that qualified as being "about abortion" (half or more of the content focused on abortion-related issues). In the U.S., we could use the newspaper's own index to lead us to appropriate articles, although we also rejected some articles that were indexed under "abortion" as having too little relevant content to qualify. Especially in the early years of our sample, we checked other related topics (thalidomide, birth defects, contraception) in the index to find articles that met our criteria. Since the NYT Index went back into the 1960s and the LAT Index only began in 1972, we selected two earlier years (1962 and 1967) from the NYT and began our continuous series from 1970 in the NYT and 1972 in the LAT.

"Sampled articles had to meet a number of other screening criteria to be in our sample. In the US, we excluded articles of less than three paragraphs, preferring to code the content-richer, longer articles. This means that in some comparisons of discourse style we also excluded the short news reports in the German sample in order to be comparing only similar types of articles. In both countries, editorials, op-ed columns, news analyses, and news accounts were included but book reviews and letters to the editor were not.

"As Table 3.1 shows, there were many more articles available about abortion in the U.S. than in Germany. Since we wanted actual sample sizes that were approximately equal in each country (we were aiming for about 1200 articles in each), we adopted different sampling strategies in each country. In the United States, we used a sampling strategy that aimed to produce a minimum of twenty articles per year with roughly equal numbers of articles from each newspaper, and would not overly concentrate our sample on the explosion of abortion-related articles that appeared around the 1989 Webster decision. This meant taking a higher proportion of articles in slow years and a lower proportion in others. It also meant that we took a somewhat larger fraction of LAT articles than those from the NYT. The New York Times index, for example, showed 85 articles on abortion in 1987 and 377 articles in 1989. We thus set a variable sampling fraction for each year and newspaper and drew our articles randomly within each subset to achieve that number. Thus, while we chose a random sample of 1 in 4 1987 articles, we sampled only 1 in 8 of the 1989 set.

"To select the articles dealing with the abortion issue in the FAZ and SZ we made use of the index of the archive of the German Bundestag. To be sure that the selection criteria of this archive were not biased, for example, in favor of parties that are members of the Bundestag, we checked six different weeks to see whether or not the articles we found in the archive were identical with the articles on abortion in the two newspapers. We did not find any bias. There were fewer articles in Germany than in the US overall, even in years such as 1990-92 when there was intense coverage. As a result, the German sample was created by taking every other article in the period 1970-1979 and every article in the 1980-1994 period.

"This oversampling of articles in slow years in the US and in recent years in Germany does not affect our comparisons of changes over time because we then weighted the data to reflect the true proportions of articles in the population. For example, the sampled articles in the first period in Germany were weighted by a factor of two. If we were to use unweighted numbers, this would impart a bias to our estimates of overall standing or frame prominence in a country by overcounting certain years or certain newspapers. But by weighting data based on the sampling fraction and then "deflating" the numbers back to the actual sample size, we both capture the actual distribution over time and preserve a fair estimate of statistical significance .

"Because the numbers of articles are so small in some years, our analysis typically groups the data into six larger and unequal periods that reflect these distributions and important turning points. This allows us enough cases for analysis in even slower news periods and also makes the time periods align comparably in the two countries, even though there are small differences within period with regard to when each country acted.

"As Chart 3.2 shows, the actual distribution of articles over time is quite different in the U.S. and in Germany. The German coverage of the abortion issue is largely concentrated in the two primary periods when reform debates were occurring in the national legislature, and coverage is very light in the intervening period. Thus, the German distribution of articles has two widely separated but roughly equal peaks. In the U.S., coverage increases over time, rising to a single, very pronounced peak in the period when the Supreme Court was considering the Webster and Casey cases. The two different distributions of the extent of news coverage over time is itself an interesting fact about the discourses of the two countries that we will discuss mainly in Chapter Eleven."