Appropriate test for quantitative content analysis of one newspaper

Hi everyone,

I am examining the differences in media coverage of homicide on the basis of the religion of the victim (bias).

The sample is one newspaper over several years.

The quasi-independent variable is religion (1 = christian, 2 = muslim).. This is not a dichotomous variable as obviously there are more than two religions.

the dependent variable is words per article (newspaper story)

Let's say after examining several years of newspapers the results are
Christian: Mean = 500 (words per article) n = 40
Muslim: Mean = 250 (words per article) n = 20
And of course there are standard deviation, variance, other statistics.

What statistical test can I use to compare these means to test for significance? On the surface I thought this was simple, but have been flip flopping around due to the nature of the sample (one newspaper) it seems to be both independent, but also repeated measures, yet can not use a repeated measures.I've spent hours through textbooks and forums but no solution.

If anyone has a clear cut solution would really appreciate detailed reasoning for the conclusion so I can understand why that solution was arrived at.


TS Contributor
I'd suggest to treat the n=60 observations as independent.

Maybe one could have a look on whether same incidents
were covered multiple times, and whether there are only a few
authors involved. And whether there is a development over
time (correlation of length with monh of observation).

How much you can generalize results to other newspapers
will be a matter of discussion.

With kind regards

Thanks Karabiner,

I was thinking along the lines of each article (covering one unique incident) as being independent, since that same incident is not repeated under a different condition, although on the other hand the same phenomenon, coverage of a specific type of incident is repeated under both conditions which confuses me to no end. It seems like a perspective thing.

I have read some articles on similar research where a correlation was used, however the authors did not specify what type. Further, since this research design's quasi-independent variable isn't a dichotomy although there are two conditions i thought it ruled it out of a correlation test!

I'm purposely keeping the sample set to 12 months because I want to minimise any effects of news room culture change over time which might affect reporting...

~ Mikkey


TS Contributor
Personally, I would look for clustering or for autocorrelation anyway. Maybe many incidents in a short time exert influence on the length of articles. Just to examine possible influences in addition to religion.

With kind regards


I am definitely exploring the idea that many incidents, or a single large or prominent issue might prime the newspaper to publish more incidents and in more depth.. creating a moral panic..

~ Mikkey