Why Opinion and Personal Observation isn’t as good as Systematic Research

Families of bilingual children with developmental language disorder (DLD) are often told to use only one language. School district personnel may insist that these children receive instruction in only one language even if there are bilingual programs available. Even bilingual personnel who work with children (teachers and SLPs for example) may say that children with DLD can become more confused if in a bilingual environment. This is simply not true. I have participated in many studies that demonstrate that bilingual children are not more likely to show higher risk for DLD than monolinguals; we know that bilingual children with DLD show comparable performance to monolingual children with DLD; we know that bilingual children with DLD show cognate advantages similar to typical bilinguals; we know that intervention in one language can carry over to the other language. This work is all supported by the data-based research (linked) and is consistent with work that other researchers are doing.

But, I often hear something along the lines of, ‘in my experience, kids do better if they are educated in only one language.’ The problem is that this seems logical on the surface, and people may actually be observing this, but it flies in the face of a multitude of research findings. We know that from an evidence perspective, opinion and personal experience are the least trustworthy source of evidence. We should not make our decisions based on this. And if there is research, research should ALWAYS beat opinion. Always (well, unless the research is thin– few studies, only case reports, poorly designed, etc.).

Why shouldn’t opinion and experience be trusted?

There are at least a couple of reasons. Two really important sources of bias are at play when you focus only on opinion and experience: ascertainment bias and confirmation bias (there are others but these are two I’ll talk about today).

Ascertainment bias is when not all cases are considered in making the analysis. In research, we work really hard to get a representative sample of the population. In my bilingual research for example, we will invite ALL children to a study, not just those in one class, or those teachers think are bilingual. We try to hit districts with lots of bilingual kids, but we test everyone we can. We screen, and we ask hour-by-hour what their language exposure and use is. We test in different districts, schools, and classes. This helps us to see everyone along the continuum of bilingualism. If we were to only see a subgroup of kids– the pattern of findings wouldn’t necessarily representative– depending on how they got into a study or an observation. In our individual clinical practice, kids may come to you in different ways, but usually we don’t see every kid in a school or a district. The sample that comes to us is pretty skewed. In a sample of a subgroup of kids on one person’s caseload, even over a few years it’s hard to know if the observed pattern is really a pattern or if this is a small series of outliers.  Making a decision on the basis of this pretty restricted sample may not be trustworthy. These are kids who for whatever reason are having (or someone perceives them to have) communication difficulties. Supports at a given school may be fabulous, or not so good. A bilingual program may or may not have a good curriculum. There may be different approaches at different grades and so one. All these factors contribute to how well bilingual children with DLD will perform educationally. Low performance could be due to DLD, bilingualism, curriculum, teacher, school, neighborhood or any number of factors. In research we try to collect extensive data on those factors and to control for them so that we can understand what is really going on. But, forming an opinion based on a small local sample where most kids aren’t tested can’t really beat larger, represenative samples. These skewed patterns can lead to wrong conclusions.

The second bias I want to talk about is confirmation bias. This is when we interpret new data as confirmation of already formed theory or opinion. The tendency is to interpret data that aligns with our beliefs as confirming a hypothesis and to disregard data that does not. In research we try to guard against confirmation bias in different ways. One way is to test kids without knowing who has DLD (or is suspected of DLD). So that we can make clean comparisons. We rotate testers, so that a kid might be tested in one language by one person and in the other language by another person. This helps us to not make assumptions about what kids know based on how they already did. On our dynamic assessment studies for example, each child was tested by 3 people– one person for pretest, one person for intervention, and a different person for posttest. This way we reduced bias. Each person worked with the child without assumptions. We video or audiotape our sessions and these are checked to make sure that procedures were followed and that everything was transcribed correctly. School SLPs don’t have the luxury of doing these things because we are responsible for ALL the testing. I think that people try to guard against confirmation bias, but it’s human nature to form a first impression.

I think that talking about these biases and realizing that they occur can help us to reduce bias in our thinking. But it is there. I think we have to sit with what our observations are in the context of research findings and think about what kinds of bias may be in play that could potentially mislead.



, , , ,

  1. Leave a comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: