#MRX MRA IMRO SMR Guidelines #4 Reliability: The Conversition Commentary

November 8, 2010 | Comments Off

MRA recently released version 1 of the MRA/IMRO Guide to the Top 16 Social Media Research Questions, a tool to help newcomers and vendors communicate with each other about this new datasource and method. Conversition was a key contributor to this document which is now available on the MRA website.
.
This blog is #4 in a series of 16, each one addressing Conversition’s viewpoint on one of the items in the guidelines. We welcome your questions and comments, and look forward to further discussions on this exciting new trend in the market research industry.
.

deanjenkins from morguefile
.
How reliable are SMR results?
.
Reliability and validity are topics of particular interest to Conversition because they are staples of the market research industry that are not always well understood by market researchers and users. These quality measures apply to all methods of research, including survey and focus group research, and for best results must be considered jointly.

To begin, reliability refers to results which can be replicated across numerous occasions. For instance, if several different people were to conduct the same research, each person would achieve the same results. Or, if the same study was conducted across several time periods, each study would each achieve the same results. No matter who or where or when, every time the study is performed, the results should be the same. (Unless of course you are conducting a pre/post or time series study where you are looking for different results each time.)

Because different survey panels have different incentive and recruitment strategies, high reliability across survey panels is generally not expected nor appropriate. The same follows for social media research. Every vendor has different methods of data collection and treatment, and they each follow more or less stringent standards and processes. But, within each social media research vendor, an acceptable level of reliability over time and among product categories should be achievable.

Now, remember that reliable results do not quality make. One can easily achieve the same wrong results over and over again by using the same bad survey or bad focus group or bad SMR over and over again. This brings us to validity, the significant other of reliability. Validity refers to results which reflect exactly what was intended to be measured.

When we ask people to select their favourite item from a list, we achieve validity when people read the entire list instead of choosing an item at the top of the list. We achieve validity when people tell us which political candidate they honestly plan to vote for instead of giving us the name of most socially desirable candidate. Of course, people are not robots and these validity issues pop up all the time, but, we have learned many research techniques to solve these problems.

Validity in social media research comes down to the treatment of data. Data quality measures must ensure that the right data is being selected for analysis. As such, data for Apple Computers must not include data for apple pie. Similarly, data for British Petroleum (BP) must not include data for Basis Points or Blood Pressure or Boston Pizza.

Data quality practices extend beyond simply gathering the right set of data. They must be applied to other data treatments as well including sentiment analysis and content analysis. Thus, sentiment analysis must distinguish between dope that is smoked illegally and dope that is hip, cool, and totally rad. And, content analysis must distinguish between the orange fruit and the orange color and the Planet Orange charity so lovingly built by the folks at ING.

For all of these purposes, validity can be evaluated with a fairly simple process.

1. Randomly select 1000 records from across different topics, dates, and data sources.
2. Score each record yourself.  a) What brand name does it reflect? b) What sentiment score does it deserve? c) What variable does it reflect?
3. Run the data through a second system whether it be your automated processes or a second person.
4. Match the two sets of results together.
5. Calculate the percentage of results that agree. a) What percentage of the data was actually about the intended brand name? b) What percentage of sentiment scores matched? c) What percentage of variables were correct?

.

The most important component of this validation work is that the two sets of data are scored blindly. In other words, I don’t know how you scored them, and you don’t know how I scored them.

Reliability and validity are essential components of all quality market research methods, including social media research. You can have one without the other, but without both you really have nothing. You need to ask your social media research provider how they address validity and reliability. Are these words essential components of their work? Do they have processes in place? Are those process grounded in solid research standards?

Go forth and inquire. It’s time.
.
.
Related links

MRA IMRO Guide #1: Advantages and Disadvantages of SMR
MRA IMRO Guide #2: Datasources of SMR
MRA IMRO Guide #3: Data Fusion and SMR
MRA IMRO Guide #4: Reliability of SMR

We're sorry, but comments are closed.

No Responses to  “#MRX MRA IMRO SMR Guidelines #4 Reliability: The Conversition Commentary”